status-go/protocol/common/message.go

853 lines
27 KiB
Go
Raw Normal View History

2020-09-01 13:27:01 +00:00
package common
2019-08-06 21:50:13 +00:00
import (
"crypto/ecdsa"
"encoding/base64"
"encoding/hex"
"encoding/json"
"errors"
"fmt"
2023-02-02 17:59:48 +00:00
"io/ioutil"
URL unfurling (initial implementation) (#3471) This is the initial implementation for the new URL unfurling requirements. The most important one is that only the message sender will pay the privacy cost for unfurling and extracting metadata from websites. Once the message is sent, the unfurled data will be stored at the protocol level and receivers will just profit and happily decode the metadata to render it. Further development of this URL unfurling capability will be mostly guided by issues created on clients. For the moment in status-mobile: https://github.com/status-im/status-mobile/labels/url-preview - https://github.com/status-im/status-mobile/issues/15918 - https://github.com/status-im/status-mobile/issues/15917 - https://github.com/status-im/status-mobile/issues/15910 - https://github.com/status-im/status-mobile/issues/15909 - https://github.com/status-im/status-mobile/issues/15908 - https://github.com/status-im/status-mobile/issues/15906 - https://github.com/status-im/status-mobile/issues/15905 ### Terminology In the code, I've tried to stick to the word "unfurl URL" to really mean the process of extracting metadata from a website, sort of lower level. I use "link preview" to mean a higher level structure which is enriched by unfurled data. "link preview" is also how designers refer to it. ### User flows 1. Carol needs to see link previews while typing in the chat input field. Notice from the diagram nothing is persisted and that status-go endpoints are essentially stateless. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_getTextURLs Server-->>Client: Normalized URLs Client->>Client: Render cached unfurled URLs Client->>Server: Unfurl non-cached URLs.\nCall wakuext_unfurlURLs Server->>Website: Fetch metadata Website-->>Server: Metadata (thumbnail URL, title, etc) Server->>Website: Fetch thumbnail Server->>Website: Fetch favicon Website-->>Server: Favicon bytes Website-->>Server: Thumbnail bytes Server->>Server: Decode & process images Server-->>Client: Unfurled data (thumbnail data URI, etc) #+end_src ``` ``` ,------. ,------. ,-------. |Client| |Server| |Website| `--+---' `--+---' `---+---' | Call wakuext_getTextURLs | | | ---------------------------------------> | | | | | Normalized URLs | | | <- - - - - - - - - - - - - - - - - - - - | | | | |----. | | | | Render cached unfurled URLs | | |<---' | | | | | | Unfurl non-cached URLs. | | | Call wakuext_unfurlURLs | | | ---------------------------------------> | | | | | | Fetch metadata | | | ------------------------------------> | | | | | Metadata (thumbnail URL, title, etc)| | | <- - - - - - - - - - - - - - - - - - | | | | | Fetch thumbnail | | | ------------------------------------> | | | | | Fetch favicon | | | ------------------------------------> | | | | | Favicon bytes | | | <- - - - - - - - - - - - - - - - - - | | | | | Thumbnail bytes | | | <- - - - - - - - - - - - - - - - - - | | | | |----. | | | | Decode & process images | | |<---' | | | | | Unfurled data (thumbnail data URI, etc)| | | <- - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,---+---. |Client| |Server| |Website| `------' `------' `-------' ``` 2. Carol sends the text message with link previews in the RPC request wakuext_sendChatMessages. status-go assumes the link previews are good because it can't and shouldn't attempt to re-unfurl them. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_sendChatMessages Server->>Server: Transform link previews to\nbe proto-marshalled Server->DB: Write link previews serialized as JSON Server-->>Client: Updated message response #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | Call wakuext_sendChatMessages| | | -----------------------------> | | | | | |----. | | | | Transform link previews to | | |<---' be proto-marshalled | | | | | | | | | Write link previews serialized as JSON| | | --------------------------------------> | | | | Updated message response | | | <- - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` 3. The message was sent over waku and persisted locally in Carol's device. She should now see the link previews in the chat history. There can be many link previews shared by other chat members, therefore it is important to serve the assets via the media server to avoid overloading the ReactNative bridge with lots of big JSON payloads containing base64 encoded data URIs (maybe this concern is meaningless for desktop). When a client is rendering messages with link previews, they will have the field linkPreviews, and the thumbnail URL will point to the local media server. ``` #+begin_src plantuml :results verbatim Client->>Server: GET /link-preview/thumbnail (media server) Server->>DB: Read from user_messages.unfurled_links Server->Server: Unmarshal JSON Server-->>Client: HTTP Content-Type: image/jpeg/etc #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | GET /link-preview/thumbnail (media server)| | | ------------------------------------------> | | | | | | Read from user_messages.unfurled_links| | | --------------------------------------> | | | | |----. | | | | Unmarshal JSON | | |<---' | | | | | HTTP Content-Type: image/jpeg/etc | | | <- - - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` ### Some limitations of the current implementation The following points will become separate issues in status-go that I'll work on over the next couple weeks. In no order of importance: - Improve how multiple links are fetched; retries on failure and testing how unfurling behaves around the timeout limits (deterministically, not by making real HTTP calls as I did). https://github.com/status-im/status-go/issues/3498 - Unfurl favicons and store them in the protobuf too. - For this PR, I added unfurling support only for websites with OpenGraph https://ogp.me/ meta tags. Other unfurlers will be implemented on demand. The next one will probably be for oEmbed https://oembed.com/, the protocol supported by YouTube, for example. - Resize and/or compress thumbnails (and favicons). Often times, thumbnails are huge for the purposes of link previews. There is already support for compressing JPEGs in status-go, but I prefer to work with compression in a separate PR because I'd like to also solve the problem for PNGs (probably convert them to JPEGs, plus compress them). This would be a safe choice for thumbnails, favicons not so much because transparency is desirable. - Editing messages is not yet supported. - I haven't coded any artificial limit on the number of previews or on the size of the thumbnail payload. This will be done in a separate issue. I have heard the ideal solution may be to split messages into smaller chunks of ~125 KiB because of libp2p, but that might be too complicated at this stage of the product (?). - Link preview deletion. - For the moment, OpenGraph metadata is extracted by requesting data for the English language (and fallback to whatever is available). In the future, we'll want to unfurl by respecting the user's local device language. Some websites, like GoDaddy, are already localized based on the device's IP, but many aren't. - The website's description text should be limited by a certain number of characters, especially because it's outside our control. Exactly how much has not been decided yet, so it'll be done separately. - URL normalization can be tricky, so I implemented only the basics to help with caching. For example, the url https://status.im and HTTPS://status.im are considered identical. Also, a URL is considered valid for unfurling if its TLD exists according to publicsuffix.EffectiveTLDPlusOne. This was essential, otherwise the default Go url.Parse approach would consider many invalid URLs valid, and thus the server would waste resources trying to unfurl the unfurleable. ### Other requirements - If the message is edited, the link previews should reflect the edited text, not the original one. This has been aligned with the design team as well. - If the website's thumbnail or the favicon can't be fetched, just ignore them. The only mandatory piece of metadata is the website's title and URL. - Link previews in clients should be generated in near real-time, that is, as the user types, previews are updated. In mobile this performs very well, and it's what other clients like WhatsApp, Telegram, and Facebook do. ### Decisions - While the user typing in the input field, the client is constantly (debounced) asking status-go to parse the text and extract normalized URLs and then the client checks if they're already in its in-memory cache. If they are, no RPC call is made. I chose this approach to achieve the best possible performance in mobile and avoid the whole RPC overhead, since the chat experience is already not smooth enough. The mobile client uses URLs as cache keys in a hashmap, i.e. if the key is present, it means the preview is readily available (naive, but good enough for now). This decision also gave me more flexibility to find the best UX at this stage of the feature. - Due to the requirement that users should be able to see independent loading indicators for each link preview, when status-go can't unfurl a URL, it doesn't return it in the response. - As an initial implementation, I added the BLOB column unfurled_links to the user_messages table. The preview data is then serialized as JSON before being stored in this column. I felt that creating a separate table and the related code for this initial PR would be inconvenient. Is that reasonable to you? Once things stabilize I can create a proper table if we want to avoid this kind of solution with serialized columns.
2023-05-18 18:43:06 +00:00
"net/url"
2023-02-02 17:59:48 +00:00
"os"
"strings"
"unicode"
"unicode/utf8"
2019-08-06 21:50:13 +00:00
"github.com/golang/protobuf/proto"
2020-09-01 13:27:01 +00:00
"github.com/status-im/markdown"
2020-09-01 10:34:28 +00:00
"github.com/status-im/markdown/ast"
2023-03-06 08:51:09 +00:00
accountJson "github.com/status-im/status-go/account/json"
"github.com/status-im/status-go/eth-node/crypto"
2020-11-25 00:34:32 +00:00
"github.com/status-im/status-go/images"
2023-02-02 17:59:48 +00:00
"github.com/status-im/status-go/protocol/audio"
"github.com/status-im/status-go/protocol/protobuf"
2019-08-06 21:50:13 +00:00
)
// QuotedMessage contains the original text of the message replied to
type QuotedMessage struct {
2022-03-16 16:28:14 +00:00
ID string `json:"id"`
ContentType int64 `json:"contentType"`
// From is a public key of the author of the message.
From string `json:"from"`
Text string `json:"text"`
ParsedText json.RawMessage `json:"parsedText,omitempty"`
2022-03-16 16:28:14 +00:00
// ImageLocalURL is the local url of the image
ImageLocalURL string `json:"image,omitempty"`
// AudioLocalURL is the local url of the audio
AudioLocalURL string `json:"audio,omitempty"`
HasSticker bool `json:"sticker,omitempty"`
// CommunityID is the id of the community advertised
CommunityID string `json:"communityId,omitempty"`
Deleted bool `json:"deleted,omitempty"`
DeletedForMe bool `json:"deletedForMe,omitempty"`
DiscordMessage *protobuf.DiscordMessage `json:"discordMessage,omitempty"`
}
type CommandState int
const (
CommandStateRequestAddressForTransaction CommandState = iota + 1
CommandStateRequestAddressForTransactionDeclined
CommandStateRequestAddressForTransactionAccepted
CommandStateRequestTransaction
CommandStateRequestTransactionDeclined
CommandStateTransactionPending
CommandStateTransactionSent
)
type ContactRequestState int
const (
ContactRequestStatePending ContactRequestState = iota + 1
ContactRequestStateAccepted
ContactRequestStateDismissed
)
type ContactVerificationState int
const (
ContactVerificationStatePending ContactVerificationState = iota + 1
ContactVerificationStateAccepted
2022-10-24 11:33:47 +00:00
ContactVerificationStateDeclined
ContactVerificationStateTrusted
ContactVerificationStateUntrustworthy
ContactVerificationStateCanceled
)
URL unfurling (initial implementation) (#3471) This is the initial implementation for the new URL unfurling requirements. The most important one is that only the message sender will pay the privacy cost for unfurling and extracting metadata from websites. Once the message is sent, the unfurled data will be stored at the protocol level and receivers will just profit and happily decode the metadata to render it. Further development of this URL unfurling capability will be mostly guided by issues created on clients. For the moment in status-mobile: https://github.com/status-im/status-mobile/labels/url-preview - https://github.com/status-im/status-mobile/issues/15918 - https://github.com/status-im/status-mobile/issues/15917 - https://github.com/status-im/status-mobile/issues/15910 - https://github.com/status-im/status-mobile/issues/15909 - https://github.com/status-im/status-mobile/issues/15908 - https://github.com/status-im/status-mobile/issues/15906 - https://github.com/status-im/status-mobile/issues/15905 ### Terminology In the code, I've tried to stick to the word "unfurl URL" to really mean the process of extracting metadata from a website, sort of lower level. I use "link preview" to mean a higher level structure which is enriched by unfurled data. "link preview" is also how designers refer to it. ### User flows 1. Carol needs to see link previews while typing in the chat input field. Notice from the diagram nothing is persisted and that status-go endpoints are essentially stateless. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_getTextURLs Server-->>Client: Normalized URLs Client->>Client: Render cached unfurled URLs Client->>Server: Unfurl non-cached URLs.\nCall wakuext_unfurlURLs Server->>Website: Fetch metadata Website-->>Server: Metadata (thumbnail URL, title, etc) Server->>Website: Fetch thumbnail Server->>Website: Fetch favicon Website-->>Server: Favicon bytes Website-->>Server: Thumbnail bytes Server->>Server: Decode & process images Server-->>Client: Unfurled data (thumbnail data URI, etc) #+end_src ``` ``` ,------. ,------. ,-------. |Client| |Server| |Website| `--+---' `--+---' `---+---' | Call wakuext_getTextURLs | | | ---------------------------------------> | | | | | Normalized URLs | | | <- - - - - - - - - - - - - - - - - - - - | | | | |----. | | | | Render cached unfurled URLs | | |<---' | | | | | | Unfurl non-cached URLs. | | | Call wakuext_unfurlURLs | | | ---------------------------------------> | | | | | | Fetch metadata | | | ------------------------------------> | | | | | Metadata (thumbnail URL, title, etc)| | | <- - - - - - - - - - - - - - - - - - | | | | | Fetch thumbnail | | | ------------------------------------> | | | | | Fetch favicon | | | ------------------------------------> | | | | | Favicon bytes | | | <- - - - - - - - - - - - - - - - - - | | | | | Thumbnail bytes | | | <- - - - - - - - - - - - - - - - - - | | | | |----. | | | | Decode & process images | | |<---' | | | | | Unfurled data (thumbnail data URI, etc)| | | <- - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,---+---. |Client| |Server| |Website| `------' `------' `-------' ``` 2. Carol sends the text message with link previews in the RPC request wakuext_sendChatMessages. status-go assumes the link previews are good because it can't and shouldn't attempt to re-unfurl them. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_sendChatMessages Server->>Server: Transform link previews to\nbe proto-marshalled Server->DB: Write link previews serialized as JSON Server-->>Client: Updated message response #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | Call wakuext_sendChatMessages| | | -----------------------------> | | | | | |----. | | | | Transform link previews to | | |<---' be proto-marshalled | | | | | | | | | Write link previews serialized as JSON| | | --------------------------------------> | | | | Updated message response | | | <- - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` 3. The message was sent over waku and persisted locally in Carol's device. She should now see the link previews in the chat history. There can be many link previews shared by other chat members, therefore it is important to serve the assets via the media server to avoid overloading the ReactNative bridge with lots of big JSON payloads containing base64 encoded data URIs (maybe this concern is meaningless for desktop). When a client is rendering messages with link previews, they will have the field linkPreviews, and the thumbnail URL will point to the local media server. ``` #+begin_src plantuml :results verbatim Client->>Server: GET /link-preview/thumbnail (media server) Server->>DB: Read from user_messages.unfurled_links Server->Server: Unmarshal JSON Server-->>Client: HTTP Content-Type: image/jpeg/etc #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | GET /link-preview/thumbnail (media server)| | | ------------------------------------------> | | | | | | Read from user_messages.unfurled_links| | | --------------------------------------> | | | | |----. | | | | Unmarshal JSON | | |<---' | | | | | HTTP Content-Type: image/jpeg/etc | | | <- - - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` ### Some limitations of the current implementation The following points will become separate issues in status-go that I'll work on over the next couple weeks. In no order of importance: - Improve how multiple links are fetched; retries on failure and testing how unfurling behaves around the timeout limits (deterministically, not by making real HTTP calls as I did). https://github.com/status-im/status-go/issues/3498 - Unfurl favicons and store them in the protobuf too. - For this PR, I added unfurling support only for websites with OpenGraph https://ogp.me/ meta tags. Other unfurlers will be implemented on demand. The next one will probably be for oEmbed https://oembed.com/, the protocol supported by YouTube, for example. - Resize and/or compress thumbnails (and favicons). Often times, thumbnails are huge for the purposes of link previews. There is already support for compressing JPEGs in status-go, but I prefer to work with compression in a separate PR because I'd like to also solve the problem for PNGs (probably convert them to JPEGs, plus compress them). This would be a safe choice for thumbnails, favicons not so much because transparency is desirable. - Editing messages is not yet supported. - I haven't coded any artificial limit on the number of previews or on the size of the thumbnail payload. This will be done in a separate issue. I have heard the ideal solution may be to split messages into smaller chunks of ~125 KiB because of libp2p, but that might be too complicated at this stage of the product (?). - Link preview deletion. - For the moment, OpenGraph metadata is extracted by requesting data for the English language (and fallback to whatever is available). In the future, we'll want to unfurl by respecting the user's local device language. Some websites, like GoDaddy, are already localized based on the device's IP, but many aren't. - The website's description text should be limited by a certain number of characters, especially because it's outside our control. Exactly how much has not been decided yet, so it'll be done separately. - URL normalization can be tricky, so I implemented only the basics to help with caching. For example, the url https://status.im and HTTPS://status.im are considered identical. Also, a URL is considered valid for unfurling if its TLD exists according to publicsuffix.EffectiveTLDPlusOne. This was essential, otherwise the default Go url.Parse approach would consider many invalid URLs valid, and thus the server would waste resources trying to unfurl the unfurleable. ### Other requirements - If the message is edited, the link previews should reflect the edited text, not the original one. This has been aligned with the design team as well. - If the website's thumbnail or the favicon can't be fetched, just ignore them. The only mandatory piece of metadata is the website's title and URL. - Link previews in clients should be generated in near real-time, that is, as the user types, previews are updated. In mobile this performs very well, and it's what other clients like WhatsApp, Telegram, and Facebook do. ### Decisions - While the user typing in the input field, the client is constantly (debounced) asking status-go to parse the text and extract normalized URLs and then the client checks if they're already in its in-memory cache. If they are, no RPC call is made. I chose this approach to achieve the best possible performance in mobile and avoid the whole RPC overhead, since the chat experience is already not smooth enough. The mobile client uses URLs as cache keys in a hashmap, i.e. if the key is present, it means the preview is readily available (naive, but good enough for now). This decision also gave me more flexibility to find the best UX at this stage of the feature. - Due to the requirement that users should be able to see independent loading indicators for each link preview, when status-go can't unfurl a URL, it doesn't return it in the response. - As an initial implementation, I added the BLOB column unfurled_links to the user_messages table. The preview data is then serialized as JSON before being stored in this column. I felt that creating a separate table and the related code for this initial PR would be inconvenient. Is that reasonable to you? Once things stabilize I can create a proper table if we want to avoid this kind of solution with serialized columns.
2023-05-18 18:43:06 +00:00
type LinkPreviewThumbnail struct {
Width int `json:"width"`
Height int `json:"height"`
// Non-empty when the thumbnail is available via the media server, i.e. after
// the chat message is sent.
URL string `json:"url,omitempty"`
// Non-empty when the thumbnail payload needs to be shared with the client,
// but before it has been persisted.
DataURI string `json:"dataUri,omitempty"`
}
type LinkPreview struct {
URL string `json:"url"`
Hostname string `json:"hostname"`
Title string `json:"title,omitempty"`
Description string `json:"description,omitempty"`
Thumbnail LinkPreviewThumbnail `json:"thumbnail,omitempty"`
}
const EveryoneMentionTag = "0x00001"
type CommandParameters struct {
// ID is the ID of the initial message
ID string `json:"id"`
// From is the address we are sending the command from
From string `json:"from"`
// Address is the address sent with the command
Address string `json:"address"`
// Contract is the contract address for ERC20 tokens
Contract string `json:"contract"`
// Value is the value as a string sent
Value string `json:"value"`
// TransactionHash is the hash of the transaction
TransactionHash string `json:"transactionHash"`
// CommandState is the state of the command
CommandState CommandState `json:"commandState"`
// The Signature of the pk-bytes+transaction-hash from the wallet
// address originating
Signature []byte `json:"signature"`
}
2021-03-25 15:15:22 +00:00
// GapParameters is the From and To indicating the missing period in chat history
type GapParameters struct {
From uint32 `json:"from,omitempty"`
To uint32 `json:"to,omitempty"`
}
func (c *CommandParameters) IsTokenTransfer() bool {
return len(c.Contract) != 0
}
const (
2021-02-23 15:47:45 +00:00
OutgoingStatusSending = "sending"
OutgoingStatusSent = "sent"
OutgoingStatusDelivered = "delivered"
)
type Messages []*Message
func (m Messages) GetClock(i int) uint64 {
return m[i].Clock
}
2019-08-06 21:50:13 +00:00
// Message represents a message record in the database,
// more specifically in user_messages table.
2019-08-06 21:50:13 +00:00
type Message struct {
protobuf.ChatMessage
2019-08-06 21:50:13 +00:00
// ID calculated as keccak256(compressedAuthorPubKey, data) where data is unencrypted payload.
ID string `json:"id"`
// WhisperTimestamp is a timestamp of a Whisper envelope.
WhisperTimestamp uint64 `json:"whisperTimestamp"`
2019-08-06 21:50:13 +00:00
// From is a public key of the author of the message.
From string `json:"from"`
// Random 3 words name
Alias string `json:"alias"`
// Identicon of the author
Identicon string `json:"identicon"`
// The chat id to be stored locally
LocalChatID string `json:"localChatId"`
2021-02-23 15:47:45 +00:00
// Seen set to true when user have read this message already
2019-08-06 21:50:13 +00:00
Seen bool `json:"seen"`
OutgoingStatus string `json:"outgoingStatus,omitempty"`
QuotedMessage *QuotedMessage `json:"quotedMessage"`
// CommandParameters is the parameters sent with the message
CommandParameters *CommandParameters `json:"commandParameters"`
2021-03-25 15:15:22 +00:00
// GapParameters is the value from/to related to the gap
GapParameters *GapParameters `json:"gapParameters,omitempty"`
// Computed fields
// RTL is whether this is a right-to-left message (arabic/hebrew script etc)
RTL bool `json:"rtl"`
// ParsedText is the parsed markdown for displaying
ParsedText []byte `json:"parsedText,omitempty"`
// ParsedTextAst is the ast of the parsed text
ParsedTextAst *ast.Node `json:"-"`
// LineCount is the count of newlines in the message
LineCount int `json:"lineCount"`
// Base64Image is the converted base64 image
Base64Image string `json:"image,omitempty"`
// ImagePath is the path of the image to be sent
ImagePath string `json:"imagePath,omitempty"`
// Base64Audio is the converted base64 audio
Base64Audio string `json:"audio,omitempty"`
// AudioPath is the path of the audio to be sent
AudioPath string `json:"audioPath,omitempty"`
// ImageLocalURL is the local url of the image
ImageLocalURL string `json:"imageLocalUrl,omitempty"`
2022-02-23 14:34:16 +00:00
// AudioLocalURL is the local url of the audio
AudioLocalURL string `json:"audioLocalUrl,omitempty"`
// StickerLocalURL is the local url of the sticker
StickerLocalURL string `json:"stickerLocalUrl,omitempty"`
// CommunityID is the id of the community to advertise
CommunityID string `json:"communityId,omitempty"`
2021-06-03 13:11:55 +00:00
// Replace indicates that this is a replacement of a message
// that has been updated
Replace string `json:"replace,omitempty"`
New bool `json:"new,omitempty"`
SigPubKey *ecdsa.PublicKey `json:"-"`
2020-09-01 10:34:28 +00:00
// Mentions is an array of mentions for a given message
Mentions []string
2020-10-27 17:35:28 +00:00
// Mentioned is whether the user is mentioned in the message
Mentioned bool `json:"mentioned"`
// Replied is whether the user is replied to in the message
Replied bool `json:"replied"`
2020-10-27 17:35:28 +00:00
// Links is an array of links within given message
URL unfurling (initial implementation) (#3471) This is the initial implementation for the new URL unfurling requirements. The most important one is that only the message sender will pay the privacy cost for unfurling and extracting metadata from websites. Once the message is sent, the unfurled data will be stored at the protocol level and receivers will just profit and happily decode the metadata to render it. Further development of this URL unfurling capability will be mostly guided by issues created on clients. For the moment in status-mobile: https://github.com/status-im/status-mobile/labels/url-preview - https://github.com/status-im/status-mobile/issues/15918 - https://github.com/status-im/status-mobile/issues/15917 - https://github.com/status-im/status-mobile/issues/15910 - https://github.com/status-im/status-mobile/issues/15909 - https://github.com/status-im/status-mobile/issues/15908 - https://github.com/status-im/status-mobile/issues/15906 - https://github.com/status-im/status-mobile/issues/15905 ### Terminology In the code, I've tried to stick to the word "unfurl URL" to really mean the process of extracting metadata from a website, sort of lower level. I use "link preview" to mean a higher level structure which is enriched by unfurled data. "link preview" is also how designers refer to it. ### User flows 1. Carol needs to see link previews while typing in the chat input field. Notice from the diagram nothing is persisted and that status-go endpoints are essentially stateless. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_getTextURLs Server-->>Client: Normalized URLs Client->>Client: Render cached unfurled URLs Client->>Server: Unfurl non-cached URLs.\nCall wakuext_unfurlURLs Server->>Website: Fetch metadata Website-->>Server: Metadata (thumbnail URL, title, etc) Server->>Website: Fetch thumbnail Server->>Website: Fetch favicon Website-->>Server: Favicon bytes Website-->>Server: Thumbnail bytes Server->>Server: Decode & process images Server-->>Client: Unfurled data (thumbnail data URI, etc) #+end_src ``` ``` ,------. ,------. ,-------. |Client| |Server| |Website| `--+---' `--+---' `---+---' | Call wakuext_getTextURLs | | | ---------------------------------------> | | | | | Normalized URLs | | | <- - - - - - - - - - - - - - - - - - - - | | | | |----. | | | | Render cached unfurled URLs | | |<---' | | | | | | Unfurl non-cached URLs. | | | Call wakuext_unfurlURLs | | | ---------------------------------------> | | | | | | Fetch metadata | | | ------------------------------------> | | | | | Metadata (thumbnail URL, title, etc)| | | <- - - - - - - - - - - - - - - - - - | | | | | Fetch thumbnail | | | ------------------------------------> | | | | | Fetch favicon | | | ------------------------------------> | | | | | Favicon bytes | | | <- - - - - - - - - - - - - - - - - - | | | | | Thumbnail bytes | | | <- - - - - - - - - - - - - - - - - - | | | | |----. | | | | Decode & process images | | |<---' | | | | | Unfurled data (thumbnail data URI, etc)| | | <- - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,---+---. |Client| |Server| |Website| `------' `------' `-------' ``` 2. Carol sends the text message with link previews in the RPC request wakuext_sendChatMessages. status-go assumes the link previews are good because it can't and shouldn't attempt to re-unfurl them. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_sendChatMessages Server->>Server: Transform link previews to\nbe proto-marshalled Server->DB: Write link previews serialized as JSON Server-->>Client: Updated message response #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | Call wakuext_sendChatMessages| | | -----------------------------> | | | | | |----. | | | | Transform link previews to | | |<---' be proto-marshalled | | | | | | | | | Write link previews serialized as JSON| | | --------------------------------------> | | | | Updated message response | | | <- - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` 3. The message was sent over waku and persisted locally in Carol's device. She should now see the link previews in the chat history. There can be many link previews shared by other chat members, therefore it is important to serve the assets via the media server to avoid overloading the ReactNative bridge with lots of big JSON payloads containing base64 encoded data URIs (maybe this concern is meaningless for desktop). When a client is rendering messages with link previews, they will have the field linkPreviews, and the thumbnail URL will point to the local media server. ``` #+begin_src plantuml :results verbatim Client->>Server: GET /link-preview/thumbnail (media server) Server->>DB: Read from user_messages.unfurled_links Server->Server: Unmarshal JSON Server-->>Client: HTTP Content-Type: image/jpeg/etc #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | GET /link-preview/thumbnail (media server)| | | ------------------------------------------> | | | | | | Read from user_messages.unfurled_links| | | --------------------------------------> | | | | |----. | | | | Unmarshal JSON | | |<---' | | | | | HTTP Content-Type: image/jpeg/etc | | | <- - - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` ### Some limitations of the current implementation The following points will become separate issues in status-go that I'll work on over the next couple weeks. In no order of importance: - Improve how multiple links are fetched; retries on failure and testing how unfurling behaves around the timeout limits (deterministically, not by making real HTTP calls as I did). https://github.com/status-im/status-go/issues/3498 - Unfurl favicons and store them in the protobuf too. - For this PR, I added unfurling support only for websites with OpenGraph https://ogp.me/ meta tags. Other unfurlers will be implemented on demand. The next one will probably be for oEmbed https://oembed.com/, the protocol supported by YouTube, for example. - Resize and/or compress thumbnails (and favicons). Often times, thumbnails are huge for the purposes of link previews. There is already support for compressing JPEGs in status-go, but I prefer to work with compression in a separate PR because I'd like to also solve the problem for PNGs (probably convert them to JPEGs, plus compress them). This would be a safe choice for thumbnails, favicons not so much because transparency is desirable. - Editing messages is not yet supported. - I haven't coded any artificial limit on the number of previews or on the size of the thumbnail payload. This will be done in a separate issue. I have heard the ideal solution may be to split messages into smaller chunks of ~125 KiB because of libp2p, but that might be too complicated at this stage of the product (?). - Link preview deletion. - For the moment, OpenGraph metadata is extracted by requesting data for the English language (and fallback to whatever is available). In the future, we'll want to unfurl by respecting the user's local device language. Some websites, like GoDaddy, are already localized based on the device's IP, but many aren't. - The website's description text should be limited by a certain number of characters, especially because it's outside our control. Exactly how much has not been decided yet, so it'll be done separately. - URL normalization can be tricky, so I implemented only the basics to help with caching. For example, the url https://status.im and HTTPS://status.im are considered identical. Also, a URL is considered valid for unfurling if its TLD exists according to publicsuffix.EffectiveTLDPlusOne. This was essential, otherwise the default Go url.Parse approach would consider many invalid URLs valid, and thus the server would waste resources trying to unfurl the unfurleable. ### Other requirements - If the message is edited, the link previews should reflect the edited text, not the original one. This has been aligned with the design team as well. - If the website's thumbnail or the favicon can't be fetched, just ignore them. The only mandatory piece of metadata is the website's title and URL. - Link previews in clients should be generated in near real-time, that is, as the user types, previews are updated. In mobile this performs very well, and it's what other clients like WhatsApp, Telegram, and Facebook do. ### Decisions - While the user typing in the input field, the client is constantly (debounced) asking status-go to parse the text and extract normalized URLs and then the client checks if they're already in its in-memory cache. If they are, no RPC call is made. I chose this approach to achieve the best possible performance in mobile and avoid the whole RPC overhead, since the chat experience is already not smooth enough. The mobile client uses URLs as cache keys in a hashmap, i.e. if the key is present, it means the preview is readily available (naive, but good enough for now). This decision also gave me more flexibility to find the best UX at this stage of the feature. - Due to the requirement that users should be able to see independent loading indicators for each link preview, when status-go can't unfurl a URL, it doesn't return it in the response. - As an initial implementation, I added the BLOB column unfurled_links to the user_messages table. The preview data is then serialized as JSON before being stored in this column. I felt that creating a separate table and the related code for this initial PR would be inconvenient. Is that reasonable to you? Once things stabilize I can create a proper table if we want to avoid this kind of solution with serialized columns.
2023-05-18 18:43:06 +00:00
Links []string
LinkPreviews []LinkPreview `json:"linkPreviews"`
2021-06-07 08:31:27 +00:00
// EditedAt indicates the clock value it was edited
EditedAt uint64 `json:"editedAt"`
// Deleted indicates if a message was deleted
Deleted bool `json:"deleted"`
DeletedBy string `json:"deletedBy,omitempty"`
2022-09-28 11:42:17 +00:00
DeletedForMe bool `json:"deletedForMe"`
// ContactRequestState is the state of the contact request message
ContactRequestState ContactRequestState `json:"contactRequestState,omitempty"`
// ContactVerificationState is the state of the identity verification process
ContactVerificationState ContactVerificationState `json:"contactVerificationState,omitempty"`
DiscordMessage *protobuf.DiscordMessage `json:"discordMessage,omitempty"`
}
func (m *Message) MarshalJSON() ([]byte, error) {
type StickerAlias struct {
Hash string `json:"hash"`
Pack int32 `json:"pack"`
URL string `json:"url"`
}
2023-03-06 08:51:09 +00:00
type MessageStructType struct {
2022-10-24 11:33:47 +00:00
ID string `json:"id"`
WhisperTimestamp uint64 `json:"whisperTimestamp"`
From string `json:"from"`
Alias string `json:"alias"`
Identicon string `json:"identicon"`
Seen bool `json:"seen"`
OutgoingStatus string `json:"outgoingStatus,omitempty"`
QuotedMessage *QuotedMessage `json:"quotedMessage"`
RTL bool `json:"rtl"`
ParsedText json.RawMessage `json:"parsedText,omitempty"`
LineCount int `json:"lineCount"`
Text string `json:"text"`
ChatID string `json:"chatId"`
LocalChatID string `json:"localChatId"`
Clock uint64 `json:"clock"`
Replace string `json:"replace"`
ResponseTo string `json:"responseTo"`
New bool `json:"new,omitempty"`
EnsName string `json:"ensName"`
DisplayName string `json:"displayName"`
Image string `json:"image,omitempty"`
AlbumID string `json:"albumId,omitempty"`
ImageWidth uint32 `json:"imageWidth,omitempty"`
ImageHeight uint32 `json:"imageHeight,omitempty"`
AlbumImagesCount uint32 `json:"albumImagesCount,omitempty"`
2022-10-24 11:33:47 +00:00
Audio string `json:"audio,omitempty"`
AudioDurationMs uint64 `json:"audioDurationMs,omitempty"`
CommunityID string `json:"communityId,omitempty"`
Sticker *StickerAlias `json:"sticker,omitempty"`
CommandParameters *CommandParameters `json:"commandParameters,omitempty"`
GapParameters *GapParameters `json:"gapParameters,omitempty"`
Timestamp uint64 `json:"timestamp"`
ContentType protobuf.ChatMessage_ContentType `json:"contentType"`
MessageType protobuf.MessageType `json:"messageType"`
Mentions []string `json:"mentions,omitempty"`
Mentioned bool `json:"mentioned,omitempty"`
Replied bool `json:"replied,omitempty"`
2022-10-24 11:33:47 +00:00
Links []string `json:"links,omitempty"`
URL unfurling (initial implementation) (#3471) This is the initial implementation for the new URL unfurling requirements. The most important one is that only the message sender will pay the privacy cost for unfurling and extracting metadata from websites. Once the message is sent, the unfurled data will be stored at the protocol level and receivers will just profit and happily decode the metadata to render it. Further development of this URL unfurling capability will be mostly guided by issues created on clients. For the moment in status-mobile: https://github.com/status-im/status-mobile/labels/url-preview - https://github.com/status-im/status-mobile/issues/15918 - https://github.com/status-im/status-mobile/issues/15917 - https://github.com/status-im/status-mobile/issues/15910 - https://github.com/status-im/status-mobile/issues/15909 - https://github.com/status-im/status-mobile/issues/15908 - https://github.com/status-im/status-mobile/issues/15906 - https://github.com/status-im/status-mobile/issues/15905 ### Terminology In the code, I've tried to stick to the word "unfurl URL" to really mean the process of extracting metadata from a website, sort of lower level. I use "link preview" to mean a higher level structure which is enriched by unfurled data. "link preview" is also how designers refer to it. ### User flows 1. Carol needs to see link previews while typing in the chat input field. Notice from the diagram nothing is persisted and that status-go endpoints are essentially stateless. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_getTextURLs Server-->>Client: Normalized URLs Client->>Client: Render cached unfurled URLs Client->>Server: Unfurl non-cached URLs.\nCall wakuext_unfurlURLs Server->>Website: Fetch metadata Website-->>Server: Metadata (thumbnail URL, title, etc) Server->>Website: Fetch thumbnail Server->>Website: Fetch favicon Website-->>Server: Favicon bytes Website-->>Server: Thumbnail bytes Server->>Server: Decode & process images Server-->>Client: Unfurled data (thumbnail data URI, etc) #+end_src ``` ``` ,------. ,------. ,-------. |Client| |Server| |Website| `--+---' `--+---' `---+---' | Call wakuext_getTextURLs | | | ---------------------------------------> | | | | | Normalized URLs | | | <- - - - - - - - - - - - - - - - - - - - | | | | |----. | | | | Render cached unfurled URLs | | |<---' | | | | | | Unfurl non-cached URLs. | | | Call wakuext_unfurlURLs | | | ---------------------------------------> | | | | | | Fetch metadata | | | ------------------------------------> | | | | | Metadata (thumbnail URL, title, etc)| | | <- - - - - - - - - - - - - - - - - - | | | | | Fetch thumbnail | | | ------------------------------------> | | | | | Fetch favicon | | | ------------------------------------> | | | | | Favicon bytes | | | <- - - - - - - - - - - - - - - - - - | | | | | Thumbnail bytes | | | <- - - - - - - - - - - - - - - - - - | | | | |----. | | | | Decode & process images | | |<---' | | | | | Unfurled data (thumbnail data URI, etc)| | | <- - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,---+---. |Client| |Server| |Website| `------' `------' `-------' ``` 2. Carol sends the text message with link previews in the RPC request wakuext_sendChatMessages. status-go assumes the link previews are good because it can't and shouldn't attempt to re-unfurl them. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_sendChatMessages Server->>Server: Transform link previews to\nbe proto-marshalled Server->DB: Write link previews serialized as JSON Server-->>Client: Updated message response #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | Call wakuext_sendChatMessages| | | -----------------------------> | | | | | |----. | | | | Transform link previews to | | |<---' be proto-marshalled | | | | | | | | | Write link previews serialized as JSON| | | --------------------------------------> | | | | Updated message response | | | <- - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` 3. The message was sent over waku and persisted locally in Carol's device. She should now see the link previews in the chat history. There can be many link previews shared by other chat members, therefore it is important to serve the assets via the media server to avoid overloading the ReactNative bridge with lots of big JSON payloads containing base64 encoded data URIs (maybe this concern is meaningless for desktop). When a client is rendering messages with link previews, they will have the field linkPreviews, and the thumbnail URL will point to the local media server. ``` #+begin_src plantuml :results verbatim Client->>Server: GET /link-preview/thumbnail (media server) Server->>DB: Read from user_messages.unfurled_links Server->Server: Unmarshal JSON Server-->>Client: HTTP Content-Type: image/jpeg/etc #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | GET /link-preview/thumbnail (media server)| | | ------------------------------------------> | | | | | | Read from user_messages.unfurled_links| | | --------------------------------------> | | | | |----. | | | | Unmarshal JSON | | |<---' | | | | | HTTP Content-Type: image/jpeg/etc | | | <- - - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` ### Some limitations of the current implementation The following points will become separate issues in status-go that I'll work on over the next couple weeks. In no order of importance: - Improve how multiple links are fetched; retries on failure and testing how unfurling behaves around the timeout limits (deterministically, not by making real HTTP calls as I did). https://github.com/status-im/status-go/issues/3498 - Unfurl favicons and store them in the protobuf too. - For this PR, I added unfurling support only for websites with OpenGraph https://ogp.me/ meta tags. Other unfurlers will be implemented on demand. The next one will probably be for oEmbed https://oembed.com/, the protocol supported by YouTube, for example. - Resize and/or compress thumbnails (and favicons). Often times, thumbnails are huge for the purposes of link previews. There is already support for compressing JPEGs in status-go, but I prefer to work with compression in a separate PR because I'd like to also solve the problem for PNGs (probably convert them to JPEGs, plus compress them). This would be a safe choice for thumbnails, favicons not so much because transparency is desirable. - Editing messages is not yet supported. - I haven't coded any artificial limit on the number of previews or on the size of the thumbnail payload. This will be done in a separate issue. I have heard the ideal solution may be to split messages into smaller chunks of ~125 KiB because of libp2p, but that might be too complicated at this stage of the product (?). - Link preview deletion. - For the moment, OpenGraph metadata is extracted by requesting data for the English language (and fallback to whatever is available). In the future, we'll want to unfurl by respecting the user's local device language. Some websites, like GoDaddy, are already localized based on the device's IP, but many aren't. - The website's description text should be limited by a certain number of characters, especially because it's outside our control. Exactly how much has not been decided yet, so it'll be done separately. - URL normalization can be tricky, so I implemented only the basics to help with caching. For example, the url https://status.im and HTTPS://status.im are considered identical. Also, a URL is considered valid for unfurling if its TLD exists according to publicsuffix.EffectiveTLDPlusOne. This was essential, otherwise the default Go url.Parse approach would consider many invalid URLs valid, and thus the server would waste resources trying to unfurl the unfurleable. ### Other requirements - If the message is edited, the link previews should reflect the edited text, not the original one. This has been aligned with the design team as well. - If the website's thumbnail or the favicon can't be fetched, just ignore them. The only mandatory piece of metadata is the website's title and URL. - Link previews in clients should be generated in near real-time, that is, as the user types, previews are updated. In mobile this performs very well, and it's what other clients like WhatsApp, Telegram, and Facebook do. ### Decisions - While the user typing in the input field, the client is constantly (debounced) asking status-go to parse the text and extract normalized URLs and then the client checks if they're already in its in-memory cache. If they are, no RPC call is made. I chose this approach to achieve the best possible performance in mobile and avoid the whole RPC overhead, since the chat experience is already not smooth enough. The mobile client uses URLs as cache keys in a hashmap, i.e. if the key is present, it means the preview is readily available (naive, but good enough for now). This decision also gave me more flexibility to find the best UX at this stage of the feature. - Due to the requirement that users should be able to see independent loading indicators for each link preview, when status-go can't unfurl a URL, it doesn't return it in the response. - As an initial implementation, I added the BLOB column unfurled_links to the user_messages table. The preview data is then serialized as JSON before being stored in this column. I felt that creating a separate table and the related code for this initial PR would be inconvenient. Is that reasonable to you? Once things stabilize I can create a proper table if we want to avoid this kind of solution with serialized columns.
2023-05-18 18:43:06 +00:00
LinkPreviews []LinkPreview `json:"linkPreviews,omitempty"`
2022-10-24 11:33:47 +00:00
EditedAt uint64 `json:"editedAt,omitempty"`
Deleted bool `json:"deleted,omitempty"`
DeletedBy string `json:"deletedBy,omitempty"`
2022-10-24 11:33:47 +00:00
DeletedForMe bool `json:"deletedForMe,omitempty"`
ContactRequestState ContactRequestState `json:"contactRequestState,omitempty"`
ContactVerificationState ContactVerificationState `json:"contactVerificationState,omitempty"`
DiscordMessage *protobuf.DiscordMessage `json:"discordMessage,omitempty"`
2023-03-06 08:51:09 +00:00
}
item := MessageStructType{
2022-10-24 11:33:47 +00:00
ID: m.ID,
WhisperTimestamp: m.WhisperTimestamp,
From: m.From,
Alias: m.Alias,
Identicon: m.Identicon,
Seen: m.Seen,
OutgoingStatus: m.OutgoingStatus,
QuotedMessage: m.QuotedMessage,
RTL: m.RTL,
ParsedText: m.ParsedText,
LineCount: m.LineCount,
Text: m.Text,
Replace: m.Replace,
ChatID: m.ChatId,
LocalChatID: m.LocalChatID,
Clock: m.Clock,
ResponseTo: m.ResponseTo,
New: m.New,
EnsName: m.EnsName,
DisplayName: m.DisplayName,
Image: m.ImageLocalURL,
Audio: m.AudioLocalURL,
CommunityID: m.CommunityID,
Timestamp: m.Timestamp,
ContentType: m.ContentType,
Mentions: m.Mentions,
Mentioned: m.Mentioned,
Replied: m.Replied,
2022-10-24 11:33:47 +00:00
Links: m.Links,
URL unfurling (initial implementation) (#3471) This is the initial implementation for the new URL unfurling requirements. The most important one is that only the message sender will pay the privacy cost for unfurling and extracting metadata from websites. Once the message is sent, the unfurled data will be stored at the protocol level and receivers will just profit and happily decode the metadata to render it. Further development of this URL unfurling capability will be mostly guided by issues created on clients. For the moment in status-mobile: https://github.com/status-im/status-mobile/labels/url-preview - https://github.com/status-im/status-mobile/issues/15918 - https://github.com/status-im/status-mobile/issues/15917 - https://github.com/status-im/status-mobile/issues/15910 - https://github.com/status-im/status-mobile/issues/15909 - https://github.com/status-im/status-mobile/issues/15908 - https://github.com/status-im/status-mobile/issues/15906 - https://github.com/status-im/status-mobile/issues/15905 ### Terminology In the code, I've tried to stick to the word "unfurl URL" to really mean the process of extracting metadata from a website, sort of lower level. I use "link preview" to mean a higher level structure which is enriched by unfurled data. "link preview" is also how designers refer to it. ### User flows 1. Carol needs to see link previews while typing in the chat input field. Notice from the diagram nothing is persisted and that status-go endpoints are essentially stateless. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_getTextURLs Server-->>Client: Normalized URLs Client->>Client: Render cached unfurled URLs Client->>Server: Unfurl non-cached URLs.\nCall wakuext_unfurlURLs Server->>Website: Fetch metadata Website-->>Server: Metadata (thumbnail URL, title, etc) Server->>Website: Fetch thumbnail Server->>Website: Fetch favicon Website-->>Server: Favicon bytes Website-->>Server: Thumbnail bytes Server->>Server: Decode & process images Server-->>Client: Unfurled data (thumbnail data URI, etc) #+end_src ``` ``` ,------. ,------. ,-------. |Client| |Server| |Website| `--+---' `--+---' `---+---' | Call wakuext_getTextURLs | | | ---------------------------------------> | | | | | Normalized URLs | | | <- - - - - - - - - - - - - - - - - - - - | | | | |----. | | | | Render cached unfurled URLs | | |<---' | | | | | | Unfurl non-cached URLs. | | | Call wakuext_unfurlURLs | | | ---------------------------------------> | | | | | | Fetch metadata | | | ------------------------------------> | | | | | Metadata (thumbnail URL, title, etc)| | | <- - - - - - - - - - - - - - - - - - | | | | | Fetch thumbnail | | | ------------------------------------> | | | | | Fetch favicon | | | ------------------------------------> | | | | | Favicon bytes | | | <- - - - - - - - - - - - - - - - - - | | | | | Thumbnail bytes | | | <- - - - - - - - - - - - - - - - - - | | | | |----. | | | | Decode & process images | | |<---' | | | | | Unfurled data (thumbnail data URI, etc)| | | <- - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,---+---. |Client| |Server| |Website| `------' `------' `-------' ``` 2. Carol sends the text message with link previews in the RPC request wakuext_sendChatMessages. status-go assumes the link previews are good because it can't and shouldn't attempt to re-unfurl them. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_sendChatMessages Server->>Server: Transform link previews to\nbe proto-marshalled Server->DB: Write link previews serialized as JSON Server-->>Client: Updated message response #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | Call wakuext_sendChatMessages| | | -----------------------------> | | | | | |----. | | | | Transform link previews to | | |<---' be proto-marshalled | | | | | | | | | Write link previews serialized as JSON| | | --------------------------------------> | | | | Updated message response | | | <- - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` 3. The message was sent over waku and persisted locally in Carol's device. She should now see the link previews in the chat history. There can be many link previews shared by other chat members, therefore it is important to serve the assets via the media server to avoid overloading the ReactNative bridge with lots of big JSON payloads containing base64 encoded data URIs (maybe this concern is meaningless for desktop). When a client is rendering messages with link previews, they will have the field linkPreviews, and the thumbnail URL will point to the local media server. ``` #+begin_src plantuml :results verbatim Client->>Server: GET /link-preview/thumbnail (media server) Server->>DB: Read from user_messages.unfurled_links Server->Server: Unmarshal JSON Server-->>Client: HTTP Content-Type: image/jpeg/etc #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | GET /link-preview/thumbnail (media server)| | | ------------------------------------------> | | | | | | Read from user_messages.unfurled_links| | | --------------------------------------> | | | | |----. | | | | Unmarshal JSON | | |<---' | | | | | HTTP Content-Type: image/jpeg/etc | | | <- - - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` ### Some limitations of the current implementation The following points will become separate issues in status-go that I'll work on over the next couple weeks. In no order of importance: - Improve how multiple links are fetched; retries on failure and testing how unfurling behaves around the timeout limits (deterministically, not by making real HTTP calls as I did). https://github.com/status-im/status-go/issues/3498 - Unfurl favicons and store them in the protobuf too. - For this PR, I added unfurling support only for websites with OpenGraph https://ogp.me/ meta tags. Other unfurlers will be implemented on demand. The next one will probably be for oEmbed https://oembed.com/, the protocol supported by YouTube, for example. - Resize and/or compress thumbnails (and favicons). Often times, thumbnails are huge for the purposes of link previews. There is already support for compressing JPEGs in status-go, but I prefer to work with compression in a separate PR because I'd like to also solve the problem for PNGs (probably convert them to JPEGs, plus compress them). This would be a safe choice for thumbnails, favicons not so much because transparency is desirable. - Editing messages is not yet supported. - I haven't coded any artificial limit on the number of previews or on the size of the thumbnail payload. This will be done in a separate issue. I have heard the ideal solution may be to split messages into smaller chunks of ~125 KiB because of libp2p, but that might be too complicated at this stage of the product (?). - Link preview deletion. - For the moment, OpenGraph metadata is extracted by requesting data for the English language (and fallback to whatever is available). In the future, we'll want to unfurl by respecting the user's local device language. Some websites, like GoDaddy, are already localized based on the device's IP, but many aren't. - The website's description text should be limited by a certain number of characters, especially because it's outside our control. Exactly how much has not been decided yet, so it'll be done separately. - URL normalization can be tricky, so I implemented only the basics to help with caching. For example, the url https://status.im and HTTPS://status.im are considered identical. Also, a URL is considered valid for unfurling if its TLD exists according to publicsuffix.EffectiveTLDPlusOne. This was essential, otherwise the default Go url.Parse approach would consider many invalid URLs valid, and thus the server would waste resources trying to unfurl the unfurleable. ### Other requirements - If the message is edited, the link previews should reflect the edited text, not the original one. This has been aligned with the design team as well. - If the website's thumbnail or the favicon can't be fetched, just ignore them. The only mandatory piece of metadata is the website's title and URL. - Link previews in clients should be generated in near real-time, that is, as the user types, previews are updated. In mobile this performs very well, and it's what other clients like WhatsApp, Telegram, and Facebook do. ### Decisions - While the user typing in the input field, the client is constantly (debounced) asking status-go to parse the text and extract normalized URLs and then the client checks if they're already in its in-memory cache. If they are, no RPC call is made. I chose this approach to achieve the best possible performance in mobile and avoid the whole RPC overhead, since the chat experience is already not smooth enough. The mobile client uses URLs as cache keys in a hashmap, i.e. if the key is present, it means the preview is readily available (naive, but good enough for now). This decision also gave me more flexibility to find the best UX at this stage of the feature. - Due to the requirement that users should be able to see independent loading indicators for each link preview, when status-go can't unfurl a URL, it doesn't return it in the response. - As an initial implementation, I added the BLOB column unfurled_links to the user_messages table. The preview data is then serialized as JSON before being stored in this column. I felt that creating a separate table and the related code for this initial PR would be inconvenient. Is that reasonable to you? Once things stabilize I can create a proper table if we want to avoid this kind of solution with serialized columns.
2023-05-18 18:43:06 +00:00
LinkPreviews: m.LinkPreviews,
2022-10-24 11:33:47 +00:00
MessageType: m.MessageType,
CommandParameters: m.CommandParameters,
GapParameters: m.GapParameters,
EditedAt: m.EditedAt,
Deleted: m.Deleted,
DeletedBy: m.DeletedBy,
2022-10-24 11:33:47 +00:00
DeletedForMe: m.DeletedForMe,
ContactRequestState: m.ContactRequestState,
ContactVerificationState: m.ContactVerificationState,
}
if sticker := m.GetSticker(); sticker != nil {
item.Sticker = &StickerAlias{
Pack: sticker.Pack,
Hash: sticker.Hash,
URL: m.StickerLocalURL,
}
}
2020-06-23 14:30:39 +00:00
if audio := m.GetAudio(); audio != nil {
item.AudioDurationMs = audio.DurationMs
}
if image := m.GetImage(); image != nil {
item.AlbumID = image.AlbumId
item.ImageWidth = image.Width
item.ImageHeight = image.Height
item.AlbumImagesCount = image.AlbumImagesCount
}
if discordMessage := m.GetDiscordMessage(); discordMessage != nil {
item.DiscordMessage = discordMessage
}
2023-03-06 08:51:09 +00:00
if item.From != "" {
ext, err := accountJson.ExtendStructWithPubKeyData(item.From, item)
if err != nil {
return nil, err
}
return json.Marshal(ext)
}
2020-06-23 14:30:39 +00:00
return json.Marshal(item)
}
func (m *Message) UnmarshalJSON(data []byte) error {
type Alias Message
aux := struct {
*Alias
ResponseTo string `json:"responseTo"`
EnsName string `json:"ensName"`
DisplayName string `json:"displayName"`
ChatID string `json:"chatId"`
Sticker *protobuf.StickerMessage `json:"sticker"`
AudioDurationMs uint64 `json:"audioDurationMs"`
ParsedText json.RawMessage `json:"parsedText"`
ContentType protobuf.ChatMessage_ContentType `json:"contentType"`
AlbumID string `json:"albumId"`
ImageWidth uint32 `json:"imageWidth"`
ImageHeight uint32 `json:"imageHeight"`
AlbumImagesCount uint32 `json:"albumImagesCount"`
From string `json:"from"`
Deleted bool `json:"deleted,omitempty"`
DeletedForMe bool `json:"deletedForMe,omitempty"`
}{
Alias: (*Alias)(m),
}
if err := json.Unmarshal(data, &aux); err != nil {
return err
}
2020-07-20 21:36:03 +00:00
if aux.ContentType == protobuf.ChatMessage_STICKER {
m.Payload = &protobuf.ChatMessage_Sticker{Sticker: aux.Sticker}
}
2020-06-23 14:30:39 +00:00
if aux.ContentType == protobuf.ChatMessage_AUDIO {
m.Payload = &protobuf.ChatMessage_Audio{
Audio: &protobuf.AudioMessage{DurationMs: aux.AudioDurationMs},
}
}
if aux.ContentType == protobuf.ChatMessage_IMAGE {
m.Payload = &protobuf.ChatMessage_Image{
Image: &protobuf.ImageMessage{
AlbumId: aux.AlbumID,
Width: aux.ImageWidth,
Height: aux.ImageHeight,
AlbumImagesCount: aux.AlbumImagesCount},
}
}
m.ResponseTo = aux.ResponseTo
m.EnsName = aux.EnsName
2022-02-17 15:13:10 +00:00
m.DisplayName = aux.DisplayName
m.ChatId = aux.ChatID
m.ContentType = aux.ContentType
m.ParsedText = aux.ParsedText
m.From = aux.From
m.Deleted = aux.Deleted
m.DeletedForMe = aux.DeletedForMe
return nil
}
// Check if the first character is Hebrew or Arabic or the RTL character
func isRTL(s string) bool {
first, _ := utf8.DecodeRuneInString(s)
return unicode.Is(unicode.Hebrew, first) ||
unicode.Is(unicode.Arabic, first) ||
// RTL character
first == '\u200f'
}
// parseImage check the message contains an image, and if so
// it creates the a base64 encoded version of it.
func (m *Message) parseImage() error {
if m.ContentType != protobuf.ChatMessage_IMAGE {
return nil
}
image := m.GetImage()
if image == nil {
return errors.New("image empty")
}
payload := image.Payload
e64 := base64.StdEncoding
maxEncLen := e64.EncodedLen(len(payload))
encBuf := make([]byte, maxEncLen)
e64.Encode(encBuf, payload)
2020-10-26 15:16:41 +00:00
mime, err := images.GetMimeType(image.Payload)
if err != nil {
return err
}
m.Base64Image = fmt.Sprintf("data:image/%s;base64,%s", mime, encBuf)
return nil
}
// parseAudio check the message contains an audio, and if so
// it creates a base64 encoded version of it.
func (m *Message) parseAudio() error {
if m.ContentType != protobuf.ChatMessage_AUDIO {
return nil
}
audio := m.GetAudio()
if audio == nil {
return errors.New("audio empty")
}
payload := audio.Payload
e64 := base64.StdEncoding
maxEncLen := e64.EncodedLen(len(payload))
encBuf := make([]byte, maxEncLen)
e64.Encode(encBuf, payload)
mime, err := getAudioMessageMIME(audio)
if err != nil {
return err
}
m.Base64Audio = fmt.Sprintf("data:audio/%s;base64,%s", mime, encBuf)
return nil
}
// implement interface of https://github.com/status-im/markdown/blob/b9fe921681227b1dace4b56364e15edb3b698308/ast/node.go#L701
type SimplifiedTextVisitor struct {
text string
canonicalNames map[string]string
}
func (v *SimplifiedTextVisitor) Visit(node ast.Node, entering bool) ast.WalkStatus {
// only on entering we fetch, otherwise we go on
if !entering {
return ast.GoToNext
}
switch n := node.(type) {
case *ast.Mention:
literal := string(n.Literal)
canonicalName, ok := v.canonicalNames[literal]
if ok {
v.text += canonicalName
} else {
v.text += literal
}
case *ast.Link:
destination := string(n.Destination)
v.text += destination
default:
var literal string
leaf := node.AsLeaf()
container := node.AsContainer()
if leaf != nil {
literal = string(leaf.Literal)
} else if container != nil {
literal = string(container.Literal)
}
v.text += literal
}
return ast.GoToNext
}
// implement interface of https://github.com/status-im/markdown/blob/b9fe921681227b1dace4b56364e15edb3b698308/ast/node.go#L701
type MentionsAndLinksVisitor struct {
identity string
mentioned bool
mentions []string
links []string
2020-09-01 10:34:28 +00:00
}
URL unfurling (initial implementation) (#3471) This is the initial implementation for the new URL unfurling requirements. The most important one is that only the message sender will pay the privacy cost for unfurling and extracting metadata from websites. Once the message is sent, the unfurled data will be stored at the protocol level and receivers will just profit and happily decode the metadata to render it. Further development of this URL unfurling capability will be mostly guided by issues created on clients. For the moment in status-mobile: https://github.com/status-im/status-mobile/labels/url-preview - https://github.com/status-im/status-mobile/issues/15918 - https://github.com/status-im/status-mobile/issues/15917 - https://github.com/status-im/status-mobile/issues/15910 - https://github.com/status-im/status-mobile/issues/15909 - https://github.com/status-im/status-mobile/issues/15908 - https://github.com/status-im/status-mobile/issues/15906 - https://github.com/status-im/status-mobile/issues/15905 ### Terminology In the code, I've tried to stick to the word "unfurl URL" to really mean the process of extracting metadata from a website, sort of lower level. I use "link preview" to mean a higher level structure which is enriched by unfurled data. "link preview" is also how designers refer to it. ### User flows 1. Carol needs to see link previews while typing in the chat input field. Notice from the diagram nothing is persisted and that status-go endpoints are essentially stateless. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_getTextURLs Server-->>Client: Normalized URLs Client->>Client: Render cached unfurled URLs Client->>Server: Unfurl non-cached URLs.\nCall wakuext_unfurlURLs Server->>Website: Fetch metadata Website-->>Server: Metadata (thumbnail URL, title, etc) Server->>Website: Fetch thumbnail Server->>Website: Fetch favicon Website-->>Server: Favicon bytes Website-->>Server: Thumbnail bytes Server->>Server: Decode & process images Server-->>Client: Unfurled data (thumbnail data URI, etc) #+end_src ``` ``` ,------. ,------. ,-------. |Client| |Server| |Website| `--+---' `--+---' `---+---' | Call wakuext_getTextURLs | | | ---------------------------------------> | | | | | Normalized URLs | | | <- - - - - - - - - - - - - - - - - - - - | | | | |----. | | | | Render cached unfurled URLs | | |<---' | | | | | | Unfurl non-cached URLs. | | | Call wakuext_unfurlURLs | | | ---------------------------------------> | | | | | | Fetch metadata | | | ------------------------------------> | | | | | Metadata (thumbnail URL, title, etc)| | | <- - - - - - - - - - - - - - - - - - | | | | | Fetch thumbnail | | | ------------------------------------> | | | | | Fetch favicon | | | ------------------------------------> | | | | | Favicon bytes | | | <- - - - - - - - - - - - - - - - - - | | | | | Thumbnail bytes | | | <- - - - - - - - - - - - - - - - - - | | | | |----. | | | | Decode & process images | | |<---' | | | | | Unfurled data (thumbnail data URI, etc)| | | <- - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,---+---. |Client| |Server| |Website| `------' `------' `-------' ``` 2. Carol sends the text message with link previews in the RPC request wakuext_sendChatMessages. status-go assumes the link previews are good because it can't and shouldn't attempt to re-unfurl them. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_sendChatMessages Server->>Server: Transform link previews to\nbe proto-marshalled Server->DB: Write link previews serialized as JSON Server-->>Client: Updated message response #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | Call wakuext_sendChatMessages| | | -----------------------------> | | | | | |----. | | | | Transform link previews to | | |<---' be proto-marshalled | | | | | | | | | Write link previews serialized as JSON| | | --------------------------------------> | | | | Updated message response | | | <- - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` 3. The message was sent over waku and persisted locally in Carol's device. She should now see the link previews in the chat history. There can be many link previews shared by other chat members, therefore it is important to serve the assets via the media server to avoid overloading the ReactNative bridge with lots of big JSON payloads containing base64 encoded data URIs (maybe this concern is meaningless for desktop). When a client is rendering messages with link previews, they will have the field linkPreviews, and the thumbnail URL will point to the local media server. ``` #+begin_src plantuml :results verbatim Client->>Server: GET /link-preview/thumbnail (media server) Server->>DB: Read from user_messages.unfurled_links Server->Server: Unmarshal JSON Server-->>Client: HTTP Content-Type: image/jpeg/etc #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | GET /link-preview/thumbnail (media server)| | | ------------------------------------------> | | | | | | Read from user_messages.unfurled_links| | | --------------------------------------> | | | | |----. | | | | Unmarshal JSON | | |<---' | | | | | HTTP Content-Type: image/jpeg/etc | | | <- - - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` ### Some limitations of the current implementation The following points will become separate issues in status-go that I'll work on over the next couple weeks. In no order of importance: - Improve how multiple links are fetched; retries on failure and testing how unfurling behaves around the timeout limits (deterministically, not by making real HTTP calls as I did). https://github.com/status-im/status-go/issues/3498 - Unfurl favicons and store them in the protobuf too. - For this PR, I added unfurling support only for websites with OpenGraph https://ogp.me/ meta tags. Other unfurlers will be implemented on demand. The next one will probably be for oEmbed https://oembed.com/, the protocol supported by YouTube, for example. - Resize and/or compress thumbnails (and favicons). Often times, thumbnails are huge for the purposes of link previews. There is already support for compressing JPEGs in status-go, but I prefer to work with compression in a separate PR because I'd like to also solve the problem for PNGs (probably convert them to JPEGs, plus compress them). This would be a safe choice for thumbnails, favicons not so much because transparency is desirable. - Editing messages is not yet supported. - I haven't coded any artificial limit on the number of previews or on the size of the thumbnail payload. This will be done in a separate issue. I have heard the ideal solution may be to split messages into smaller chunks of ~125 KiB because of libp2p, but that might be too complicated at this stage of the product (?). - Link preview deletion. - For the moment, OpenGraph metadata is extracted by requesting data for the English language (and fallback to whatever is available). In the future, we'll want to unfurl by respecting the user's local device language. Some websites, like GoDaddy, are already localized based on the device's IP, but many aren't. - The website's description text should be limited by a certain number of characters, especially because it's outside our control. Exactly how much has not been decided yet, so it'll be done separately. - URL normalization can be tricky, so I implemented only the basics to help with caching. For example, the url https://status.im and HTTPS://status.im are considered identical. Also, a URL is considered valid for unfurling if its TLD exists according to publicsuffix.EffectiveTLDPlusOne. This was essential, otherwise the default Go url.Parse approach would consider many invalid URLs valid, and thus the server would waste resources trying to unfurl the unfurleable. ### Other requirements - If the message is edited, the link previews should reflect the edited text, not the original one. This has been aligned with the design team as well. - If the website's thumbnail or the favicon can't be fetched, just ignore them. The only mandatory piece of metadata is the website's title and URL. - Link previews in clients should be generated in near real-time, that is, as the user types, previews are updated. In mobile this performs very well, and it's what other clients like WhatsApp, Telegram, and Facebook do. ### Decisions - While the user typing in the input field, the client is constantly (debounced) asking status-go to parse the text and extract normalized URLs and then the client checks if they're already in its in-memory cache. If they are, no RPC call is made. I chose this approach to achieve the best possible performance in mobile and avoid the whole RPC overhead, since the chat experience is already not smooth enough. The mobile client uses URLs as cache keys in a hashmap, i.e. if the key is present, it means the preview is readily available (naive, but good enough for now). This decision also gave me more flexibility to find the best UX at this stage of the feature. - Due to the requirement that users should be able to see independent loading indicators for each link preview, when status-go can't unfurl a URL, it doesn't return it in the response. - As an initial implementation, I added the BLOB column unfurled_links to the user_messages table. The preview data is then serialized as JSON before being stored in this column. I felt that creating a separate table and the related code for this initial PR would be inconvenient. Is that reasonable to you? Once things stabilize I can create a proper table if we want to avoid this kind of solution with serialized columns.
2023-05-18 18:43:06 +00:00
type LinksVisitor struct {
Links []string
}
func (v *MentionsAndLinksVisitor) Visit(node ast.Node, entering bool) ast.WalkStatus {
2020-09-01 10:34:28 +00:00
// only on entering we fetch, otherwise we go on
if !entering {
return ast.GoToNext
}
switch n := node.(type) {
case *ast.Mention:
mention := string(n.Literal)
if mention == v.identity || mention == EveryoneMentionTag {
v.mentioned = true
}
v.mentions = append(v.mentions, mention)
2020-10-27 17:35:28 +00:00
case *ast.Link:
v.links = append(v.links, string(n.Destination))
2020-09-01 10:34:28 +00:00
}
2020-10-27 17:35:28 +00:00
2020-09-01 10:34:28 +00:00
return ast.GoToNext
}
URL unfurling (initial implementation) (#3471) This is the initial implementation for the new URL unfurling requirements. The most important one is that only the message sender will pay the privacy cost for unfurling and extracting metadata from websites. Once the message is sent, the unfurled data will be stored at the protocol level and receivers will just profit and happily decode the metadata to render it. Further development of this URL unfurling capability will be mostly guided by issues created on clients. For the moment in status-mobile: https://github.com/status-im/status-mobile/labels/url-preview - https://github.com/status-im/status-mobile/issues/15918 - https://github.com/status-im/status-mobile/issues/15917 - https://github.com/status-im/status-mobile/issues/15910 - https://github.com/status-im/status-mobile/issues/15909 - https://github.com/status-im/status-mobile/issues/15908 - https://github.com/status-im/status-mobile/issues/15906 - https://github.com/status-im/status-mobile/issues/15905 ### Terminology In the code, I've tried to stick to the word "unfurl URL" to really mean the process of extracting metadata from a website, sort of lower level. I use "link preview" to mean a higher level structure which is enriched by unfurled data. "link preview" is also how designers refer to it. ### User flows 1. Carol needs to see link previews while typing in the chat input field. Notice from the diagram nothing is persisted and that status-go endpoints are essentially stateless. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_getTextURLs Server-->>Client: Normalized URLs Client->>Client: Render cached unfurled URLs Client->>Server: Unfurl non-cached URLs.\nCall wakuext_unfurlURLs Server->>Website: Fetch metadata Website-->>Server: Metadata (thumbnail URL, title, etc) Server->>Website: Fetch thumbnail Server->>Website: Fetch favicon Website-->>Server: Favicon bytes Website-->>Server: Thumbnail bytes Server->>Server: Decode & process images Server-->>Client: Unfurled data (thumbnail data URI, etc) #+end_src ``` ``` ,------. ,------. ,-------. |Client| |Server| |Website| `--+---' `--+---' `---+---' | Call wakuext_getTextURLs | | | ---------------------------------------> | | | | | Normalized URLs | | | <- - - - - - - - - - - - - - - - - - - - | | | | |----. | | | | Render cached unfurled URLs | | |<---' | | | | | | Unfurl non-cached URLs. | | | Call wakuext_unfurlURLs | | | ---------------------------------------> | | | | | | Fetch metadata | | | ------------------------------------> | | | | | Metadata (thumbnail URL, title, etc)| | | <- - - - - - - - - - - - - - - - - - | | | | | Fetch thumbnail | | | ------------------------------------> | | | | | Fetch favicon | | | ------------------------------------> | | | | | Favicon bytes | | | <- - - - - - - - - - - - - - - - - - | | | | | Thumbnail bytes | | | <- - - - - - - - - - - - - - - - - - | | | | |----. | | | | Decode & process images | | |<---' | | | | | Unfurled data (thumbnail data URI, etc)| | | <- - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,---+---. |Client| |Server| |Website| `------' `------' `-------' ``` 2. Carol sends the text message with link previews in the RPC request wakuext_sendChatMessages. status-go assumes the link previews are good because it can't and shouldn't attempt to re-unfurl them. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_sendChatMessages Server->>Server: Transform link previews to\nbe proto-marshalled Server->DB: Write link previews serialized as JSON Server-->>Client: Updated message response #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | Call wakuext_sendChatMessages| | | -----------------------------> | | | | | |----. | | | | Transform link previews to | | |<---' be proto-marshalled | | | | | | | | | Write link previews serialized as JSON| | | --------------------------------------> | | | | Updated message response | | | <- - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` 3. The message was sent over waku and persisted locally in Carol's device. She should now see the link previews in the chat history. There can be many link previews shared by other chat members, therefore it is important to serve the assets via the media server to avoid overloading the ReactNative bridge with lots of big JSON payloads containing base64 encoded data URIs (maybe this concern is meaningless for desktop). When a client is rendering messages with link previews, they will have the field linkPreviews, and the thumbnail URL will point to the local media server. ``` #+begin_src plantuml :results verbatim Client->>Server: GET /link-preview/thumbnail (media server) Server->>DB: Read from user_messages.unfurled_links Server->Server: Unmarshal JSON Server-->>Client: HTTP Content-Type: image/jpeg/etc #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | GET /link-preview/thumbnail (media server)| | | ------------------------------------------> | | | | | | Read from user_messages.unfurled_links| | | --------------------------------------> | | | | |----. | | | | Unmarshal JSON | | |<---' | | | | | HTTP Content-Type: image/jpeg/etc | | | <- - - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` ### Some limitations of the current implementation The following points will become separate issues in status-go that I'll work on over the next couple weeks. In no order of importance: - Improve how multiple links are fetched; retries on failure and testing how unfurling behaves around the timeout limits (deterministically, not by making real HTTP calls as I did). https://github.com/status-im/status-go/issues/3498 - Unfurl favicons and store them in the protobuf too. - For this PR, I added unfurling support only for websites with OpenGraph https://ogp.me/ meta tags. Other unfurlers will be implemented on demand. The next one will probably be for oEmbed https://oembed.com/, the protocol supported by YouTube, for example. - Resize and/or compress thumbnails (and favicons). Often times, thumbnails are huge for the purposes of link previews. There is already support for compressing JPEGs in status-go, but I prefer to work with compression in a separate PR because I'd like to also solve the problem for PNGs (probably convert them to JPEGs, plus compress them). This would be a safe choice for thumbnails, favicons not so much because transparency is desirable. - Editing messages is not yet supported. - I haven't coded any artificial limit on the number of previews or on the size of the thumbnail payload. This will be done in a separate issue. I have heard the ideal solution may be to split messages into smaller chunks of ~125 KiB because of libp2p, but that might be too complicated at this stage of the product (?). - Link preview deletion. - For the moment, OpenGraph metadata is extracted by requesting data for the English language (and fallback to whatever is available). In the future, we'll want to unfurl by respecting the user's local device language. Some websites, like GoDaddy, are already localized based on the device's IP, but many aren't. - The website's description text should be limited by a certain number of characters, especially because it's outside our control. Exactly how much has not been decided yet, so it'll be done separately. - URL normalization can be tricky, so I implemented only the basics to help with caching. For example, the url https://status.im and HTTPS://status.im are considered identical. Also, a URL is considered valid for unfurling if its TLD exists according to publicsuffix.EffectiveTLDPlusOne. This was essential, otherwise the default Go url.Parse approach would consider many invalid URLs valid, and thus the server would waste resources trying to unfurl the unfurleable. ### Other requirements - If the message is edited, the link previews should reflect the edited text, not the original one. This has been aligned with the design team as well. - If the website's thumbnail or the favicon can't be fetched, just ignore them. The only mandatory piece of metadata is the website's title and URL. - Link previews in clients should be generated in near real-time, that is, as the user types, previews are updated. In mobile this performs very well, and it's what other clients like WhatsApp, Telegram, and Facebook do. ### Decisions - While the user typing in the input field, the client is constantly (debounced) asking status-go to parse the text and extract normalized URLs and then the client checks if they're already in its in-memory cache. If they are, no RPC call is made. I chose this approach to achieve the best possible performance in mobile and avoid the whole RPC overhead, since the chat experience is already not smooth enough. The mobile client uses URLs as cache keys in a hashmap, i.e. if the key is present, it means the preview is readily available (naive, but good enough for now). This decision also gave me more flexibility to find the best UX at this stage of the feature. - Due to the requirement that users should be able to see independent loading indicators for each link preview, when status-go can't unfurl a URL, it doesn't return it in the response. - As an initial implementation, I added the BLOB column unfurled_links to the user_messages table. The preview data is then serialized as JSON before being stored in this column. I felt that creating a separate table and the related code for this initial PR would be inconvenient. Is that reasonable to you? Once things stabilize I can create a proper table if we want to avoid this kind of solution with serialized columns.
2023-05-18 18:43:06 +00:00
func (v *LinksVisitor) Visit(node ast.Node, entering bool) ast.WalkStatus {
if !entering {
return ast.GoToNext
}
switch n := node.(type) {
case *ast.Link:
v.Links = append(v.Links, string(n.Destination))
}
return ast.GoToNext
}
func runMentionsAndLinksVisitor(parsedText ast.Node, identity string) *MentionsAndLinksVisitor {
visitor := &MentionsAndLinksVisitor{identity: identity}
2020-09-01 10:34:28 +00:00
ast.Walk(parsedText, visitor)
return visitor
2020-09-01 10:34:28 +00:00
}
URL unfurling (initial implementation) (#3471) This is the initial implementation for the new URL unfurling requirements. The most important one is that only the message sender will pay the privacy cost for unfurling and extracting metadata from websites. Once the message is sent, the unfurled data will be stored at the protocol level and receivers will just profit and happily decode the metadata to render it. Further development of this URL unfurling capability will be mostly guided by issues created on clients. For the moment in status-mobile: https://github.com/status-im/status-mobile/labels/url-preview - https://github.com/status-im/status-mobile/issues/15918 - https://github.com/status-im/status-mobile/issues/15917 - https://github.com/status-im/status-mobile/issues/15910 - https://github.com/status-im/status-mobile/issues/15909 - https://github.com/status-im/status-mobile/issues/15908 - https://github.com/status-im/status-mobile/issues/15906 - https://github.com/status-im/status-mobile/issues/15905 ### Terminology In the code, I've tried to stick to the word "unfurl URL" to really mean the process of extracting metadata from a website, sort of lower level. I use "link preview" to mean a higher level structure which is enriched by unfurled data. "link preview" is also how designers refer to it. ### User flows 1. Carol needs to see link previews while typing in the chat input field. Notice from the diagram nothing is persisted and that status-go endpoints are essentially stateless. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_getTextURLs Server-->>Client: Normalized URLs Client->>Client: Render cached unfurled URLs Client->>Server: Unfurl non-cached URLs.\nCall wakuext_unfurlURLs Server->>Website: Fetch metadata Website-->>Server: Metadata (thumbnail URL, title, etc) Server->>Website: Fetch thumbnail Server->>Website: Fetch favicon Website-->>Server: Favicon bytes Website-->>Server: Thumbnail bytes Server->>Server: Decode & process images Server-->>Client: Unfurled data (thumbnail data URI, etc) #+end_src ``` ``` ,------. ,------. ,-------. |Client| |Server| |Website| `--+---' `--+---' `---+---' | Call wakuext_getTextURLs | | | ---------------------------------------> | | | | | Normalized URLs | | | <- - - - - - - - - - - - - - - - - - - - | | | | |----. | | | | Render cached unfurled URLs | | |<---' | | | | | | Unfurl non-cached URLs. | | | Call wakuext_unfurlURLs | | | ---------------------------------------> | | | | | | Fetch metadata | | | ------------------------------------> | | | | | Metadata (thumbnail URL, title, etc)| | | <- - - - - - - - - - - - - - - - - - | | | | | Fetch thumbnail | | | ------------------------------------> | | | | | Fetch favicon | | | ------------------------------------> | | | | | Favicon bytes | | | <- - - - - - - - - - - - - - - - - - | | | | | Thumbnail bytes | | | <- - - - - - - - - - - - - - - - - - | | | | |----. | | | | Decode & process images | | |<---' | | | | | Unfurled data (thumbnail data URI, etc)| | | <- - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,---+---. |Client| |Server| |Website| `------' `------' `-------' ``` 2. Carol sends the text message with link previews in the RPC request wakuext_sendChatMessages. status-go assumes the link previews are good because it can't and shouldn't attempt to re-unfurl them. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_sendChatMessages Server->>Server: Transform link previews to\nbe proto-marshalled Server->DB: Write link previews serialized as JSON Server-->>Client: Updated message response #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | Call wakuext_sendChatMessages| | | -----------------------------> | | | | | |----. | | | | Transform link previews to | | |<---' be proto-marshalled | | | | | | | | | Write link previews serialized as JSON| | | --------------------------------------> | | | | Updated message response | | | <- - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` 3. The message was sent over waku and persisted locally in Carol's device. She should now see the link previews in the chat history. There can be many link previews shared by other chat members, therefore it is important to serve the assets via the media server to avoid overloading the ReactNative bridge with lots of big JSON payloads containing base64 encoded data URIs (maybe this concern is meaningless for desktop). When a client is rendering messages with link previews, they will have the field linkPreviews, and the thumbnail URL will point to the local media server. ``` #+begin_src plantuml :results verbatim Client->>Server: GET /link-preview/thumbnail (media server) Server->>DB: Read from user_messages.unfurled_links Server->Server: Unmarshal JSON Server-->>Client: HTTP Content-Type: image/jpeg/etc #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | GET /link-preview/thumbnail (media server)| | | ------------------------------------------> | | | | | | Read from user_messages.unfurled_links| | | --------------------------------------> | | | | |----. | | | | Unmarshal JSON | | |<---' | | | | | HTTP Content-Type: image/jpeg/etc | | | <- - - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` ### Some limitations of the current implementation The following points will become separate issues in status-go that I'll work on over the next couple weeks. In no order of importance: - Improve how multiple links are fetched; retries on failure and testing how unfurling behaves around the timeout limits (deterministically, not by making real HTTP calls as I did). https://github.com/status-im/status-go/issues/3498 - Unfurl favicons and store them in the protobuf too. - For this PR, I added unfurling support only for websites with OpenGraph https://ogp.me/ meta tags. Other unfurlers will be implemented on demand. The next one will probably be for oEmbed https://oembed.com/, the protocol supported by YouTube, for example. - Resize and/or compress thumbnails (and favicons). Often times, thumbnails are huge for the purposes of link previews. There is already support for compressing JPEGs in status-go, but I prefer to work with compression in a separate PR because I'd like to also solve the problem for PNGs (probably convert them to JPEGs, plus compress them). This would be a safe choice for thumbnails, favicons not so much because transparency is desirable. - Editing messages is not yet supported. - I haven't coded any artificial limit on the number of previews or on the size of the thumbnail payload. This will be done in a separate issue. I have heard the ideal solution may be to split messages into smaller chunks of ~125 KiB because of libp2p, but that might be too complicated at this stage of the product (?). - Link preview deletion. - For the moment, OpenGraph metadata is extracted by requesting data for the English language (and fallback to whatever is available). In the future, we'll want to unfurl by respecting the user's local device language. Some websites, like GoDaddy, are already localized based on the device's IP, but many aren't. - The website's description text should be limited by a certain number of characters, especially because it's outside our control. Exactly how much has not been decided yet, so it'll be done separately. - URL normalization can be tricky, so I implemented only the basics to help with caching. For example, the url https://status.im and HTTPS://status.im are considered identical. Also, a URL is considered valid for unfurling if its TLD exists according to publicsuffix.EffectiveTLDPlusOne. This was essential, otherwise the default Go url.Parse approach would consider many invalid URLs valid, and thus the server would waste resources trying to unfurl the unfurleable. ### Other requirements - If the message is edited, the link previews should reflect the edited text, not the original one. This has been aligned with the design team as well. - If the website's thumbnail or the favicon can't be fetched, just ignore them. The only mandatory piece of metadata is the website's title and URL. - Link previews in clients should be generated in near real-time, that is, as the user types, previews are updated. In mobile this performs very well, and it's what other clients like WhatsApp, Telegram, and Facebook do. ### Decisions - While the user typing in the input field, the client is constantly (debounced) asking status-go to parse the text and extract normalized URLs and then the client checks if they're already in its in-memory cache. If they are, no RPC call is made. I chose this approach to achieve the best possible performance in mobile and avoid the whole RPC overhead, since the chat experience is already not smooth enough. The mobile client uses URLs as cache keys in a hashmap, i.e. if the key is present, it means the preview is readily available (naive, but good enough for now). This decision also gave me more flexibility to find the best UX at this stage of the feature. - Due to the requirement that users should be able to see independent loading indicators for each link preview, when status-go can't unfurl a URL, it doesn't return it in the response. - As an initial implementation, I added the BLOB column unfurled_links to the user_messages table. The preview data is then serialized as JSON before being stored in this column. I felt that creating a separate table and the related code for this initial PR would be inconvenient. Is that reasonable to you? Once things stabilize I can create a proper table if we want to avoid this kind of solution with serialized columns.
2023-05-18 18:43:06 +00:00
func RunLinksVisitor(parsedText ast.Node) *LinksVisitor {
visitor := &LinksVisitor{}
ast.Walk(parsedText, visitor)
return visitor
}
// PrepareContent return the parsed content of the message, the line-count and whether
// is a right-to-left message
func (m *Message) PrepareContent(identity string) error {
var parsedText ast.Node
switch m.ContentType {
case protobuf.ChatMessage_DISCORD_MESSAGE:
parsedText = markdown.Parse([]byte(m.GetDiscordMessage().Content), nil)
default:
parsedText = markdown.Parse([]byte(m.Text), nil)
}
visitor := runMentionsAndLinksVisitor(parsedText, identity)
m.Mentions = visitor.mentions
m.Links = visitor.links
// Leave it set if already set, as sometimes we might run this without
// an identity
if !m.Mentioned || identity != "" {
m.Mentioned = visitor.mentioned
}
jsonParsedText, err := json.Marshal(parsedText)
if err != nil {
return err
}
m.ParsedTextAst = &parsedText
m.ParsedText = jsonParsedText
m.LineCount = strings.Count(m.Text, "\n")
m.RTL = isRTL(m.Text)
if err := m.parseImage(); err != nil {
return err
}
return m.parseAudio()
2019-08-06 21:50:13 +00:00
}
// GetSimplifiedText returns a the text stripped of all the markdown and with mentions
// replaced by canonical names
func (m *Message) GetSimplifiedText(identity string, canonicalNames map[string]string) (string, error) {
if m.ContentType == protobuf.ChatMessage_AUDIO {
return "Audio", nil
}
if m.ContentType == protobuf.ChatMessage_STICKER {
return "Sticker", nil
}
if m.ContentType == protobuf.ChatMessage_IMAGE {
return "Image", nil
}
if m.ContentType == protobuf.ChatMessage_COMMUNITY {
return "Community", nil
}
if m.ContentType == protobuf.ChatMessage_SYSTEM_MESSAGE_CONTENT_PRIVATE_GROUP {
return "Group", nil
}
if m.ParsedTextAst == nil {
err := m.PrepareContent(identity)
if err != nil {
return "", err
}
}
visitor := &SimplifiedTextVisitor{canonicalNames: canonicalNames}
ast.Walk(*m.ParsedTextAst, visitor)
return visitor.text, nil
}
func getAudioMessageMIME(i *protobuf.AudioMessage) (string, error) {
switch i.Type {
case protobuf.AudioMessage_AAC:
return "aac", nil
case protobuf.AudioMessage_AMR:
return "amr", nil
}
return "", errors.New("audio format not supported")
}
// GetSigPubKey returns an ecdsa encoded public key
// this function is required to implement the ChatEntity interface
func (m Message) GetSigPubKey() *ecdsa.PublicKey {
return m.SigPubKey
}
// GetProtoBuf returns the struct's embedded protobuf struct
// this function is required to implement the ChatEntity interface
func (m *Message) GetProtobuf() proto.Message {
return &m.ChatMessage
}
// SetMessageType a setter for the MessageType field
// this function is required to implement the ChatEntity interface
func (m *Message) SetMessageType(messageType protobuf.MessageType) {
m.MessageType = messageType
}
// WrapGroupMessage indicates whether we should wrap this in membership information
func (m *Message) WrapGroupMessage() bool {
return true
}
// GetPublicKey attempts to return or recreate the *ecdsa.PublicKey of the Message sender.
// If the m.SigPubKey is set this will be returned
// If the m.From is present the string is decoded and unmarshalled into a *ecdsa.PublicKey, the m.SigPubKey is set and returned
// Else an error is thrown
// This function differs from GetSigPubKey() as this function may return an error
func (m *Message) GetSenderPubKey() (*ecdsa.PublicKey, error) {
// TODO requires tests
if m.SigPubKey != nil {
return m.SigPubKey, nil
}
if len(m.From) > 0 {
fromB, err := hex.DecodeString(m.From[2:])
if err != nil {
return nil, err
}
senderPubKey, err := crypto.UnmarshalPubkey(fromB)
if err != nil {
return nil, err
}
m.SigPubKey = senderPubKey
return senderPubKey, nil
}
return nil, errors.New("no Message.SigPubKey or Message.From set unable to get public key")
}
2023-02-02 17:59:48 +00:00
func (m *Message) LoadAudio() error {
file, err := os.Open(m.AudioPath)
if err != nil {
return err
}
defer file.Close()
payload, err := ioutil.ReadAll(file)
if err != nil {
return err
}
audioMessage := m.GetAudio()
if audioMessage == nil {
return errors.New("no audio has been passed")
}
audioMessage.Payload = payload
audioMessage.Type = audio.Type(payload)
m.Payload = &protobuf.ChatMessage_Audio{Audio: audioMessage}
return os.Remove(m.AudioPath)
}
func (m *Message) LoadImage() error {
payload, err := images.OpenAndAdjustImage(images.CroppedImage{ImagePath: m.ImagePath}, false)
if err != nil {
return err
}
imageMessage := m.GetImage()
imageMessage.Payload = payload
imageMessage.Type = images.GetProtobufImageType(payload)
m.Payload = &protobuf.ChatMessage_Image{Image: imageMessage}
return nil
}
URL unfurling (initial implementation) (#3471) This is the initial implementation for the new URL unfurling requirements. The most important one is that only the message sender will pay the privacy cost for unfurling and extracting metadata from websites. Once the message is sent, the unfurled data will be stored at the protocol level and receivers will just profit and happily decode the metadata to render it. Further development of this URL unfurling capability will be mostly guided by issues created on clients. For the moment in status-mobile: https://github.com/status-im/status-mobile/labels/url-preview - https://github.com/status-im/status-mobile/issues/15918 - https://github.com/status-im/status-mobile/issues/15917 - https://github.com/status-im/status-mobile/issues/15910 - https://github.com/status-im/status-mobile/issues/15909 - https://github.com/status-im/status-mobile/issues/15908 - https://github.com/status-im/status-mobile/issues/15906 - https://github.com/status-im/status-mobile/issues/15905 ### Terminology In the code, I've tried to stick to the word "unfurl URL" to really mean the process of extracting metadata from a website, sort of lower level. I use "link preview" to mean a higher level structure which is enriched by unfurled data. "link preview" is also how designers refer to it. ### User flows 1. Carol needs to see link previews while typing in the chat input field. Notice from the diagram nothing is persisted and that status-go endpoints are essentially stateless. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_getTextURLs Server-->>Client: Normalized URLs Client->>Client: Render cached unfurled URLs Client->>Server: Unfurl non-cached URLs.\nCall wakuext_unfurlURLs Server->>Website: Fetch metadata Website-->>Server: Metadata (thumbnail URL, title, etc) Server->>Website: Fetch thumbnail Server->>Website: Fetch favicon Website-->>Server: Favicon bytes Website-->>Server: Thumbnail bytes Server->>Server: Decode & process images Server-->>Client: Unfurled data (thumbnail data URI, etc) #+end_src ``` ``` ,------. ,------. ,-------. |Client| |Server| |Website| `--+---' `--+---' `---+---' | Call wakuext_getTextURLs | | | ---------------------------------------> | | | | | Normalized URLs | | | <- - - - - - - - - - - - - - - - - - - - | | | | |----. | | | | Render cached unfurled URLs | | |<---' | | | | | | Unfurl non-cached URLs. | | | Call wakuext_unfurlURLs | | | ---------------------------------------> | | | | | | Fetch metadata | | | ------------------------------------> | | | | | Metadata (thumbnail URL, title, etc)| | | <- - - - - - - - - - - - - - - - - - | | | | | Fetch thumbnail | | | ------------------------------------> | | | | | Fetch favicon | | | ------------------------------------> | | | | | Favicon bytes | | | <- - - - - - - - - - - - - - - - - - | | | | | Thumbnail bytes | | | <- - - - - - - - - - - - - - - - - - | | | | |----. | | | | Decode & process images | | |<---' | | | | | Unfurled data (thumbnail data URI, etc)| | | <- - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,---+---. |Client| |Server| |Website| `------' `------' `-------' ``` 2. Carol sends the text message with link previews in the RPC request wakuext_sendChatMessages. status-go assumes the link previews are good because it can't and shouldn't attempt to re-unfurl them. ``` #+begin_src plantuml :results verbatim Client->>Server: Call wakuext_sendChatMessages Server->>Server: Transform link previews to\nbe proto-marshalled Server->DB: Write link previews serialized as JSON Server-->>Client: Updated message response #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | Call wakuext_sendChatMessages| | | -----------------------------> | | | | | |----. | | | | Transform link previews to | | |<---' be proto-marshalled | | | | | | | | | Write link previews serialized as JSON| | | --------------------------------------> | | | | Updated message response | | | <- - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` 3. The message was sent over waku and persisted locally in Carol's device. She should now see the link previews in the chat history. There can be many link previews shared by other chat members, therefore it is important to serve the assets via the media server to avoid overloading the ReactNative bridge with lots of big JSON payloads containing base64 encoded data URIs (maybe this concern is meaningless for desktop). When a client is rendering messages with link previews, they will have the field linkPreviews, and the thumbnail URL will point to the local media server. ``` #+begin_src plantuml :results verbatim Client->>Server: GET /link-preview/thumbnail (media server) Server->>DB: Read from user_messages.unfurled_links Server->Server: Unmarshal JSON Server-->>Client: HTTP Content-Type: image/jpeg/etc #+end_src ``` ``` ,------. ,------. ,--. |Client| |Server| |DB| `--+---' `--+---' `+-' | GET /link-preview/thumbnail (media server)| | | ------------------------------------------> | | | | | | Read from user_messages.unfurled_links| | | --------------------------------------> | | | | |----. | | | | Unmarshal JSON | | |<---' | | | | | HTTP Content-Type: image/jpeg/etc | | | <- - - - - - - - - - - - - - - - - - - - - | ,--+---. ,--+---. ,+-. |Client| |Server| |DB| `------' `------' `--' ``` ### Some limitations of the current implementation The following points will become separate issues in status-go that I'll work on over the next couple weeks. In no order of importance: - Improve how multiple links are fetched; retries on failure and testing how unfurling behaves around the timeout limits (deterministically, not by making real HTTP calls as I did). https://github.com/status-im/status-go/issues/3498 - Unfurl favicons and store them in the protobuf too. - For this PR, I added unfurling support only for websites with OpenGraph https://ogp.me/ meta tags. Other unfurlers will be implemented on demand. The next one will probably be for oEmbed https://oembed.com/, the protocol supported by YouTube, for example. - Resize and/or compress thumbnails (and favicons). Often times, thumbnails are huge for the purposes of link previews. There is already support for compressing JPEGs in status-go, but I prefer to work with compression in a separate PR because I'd like to also solve the problem for PNGs (probably convert them to JPEGs, plus compress them). This would be a safe choice for thumbnails, favicons not so much because transparency is desirable. - Editing messages is not yet supported. - I haven't coded any artificial limit on the number of previews or on the size of the thumbnail payload. This will be done in a separate issue. I have heard the ideal solution may be to split messages into smaller chunks of ~125 KiB because of libp2p, but that might be too complicated at this stage of the product (?). - Link preview deletion. - For the moment, OpenGraph metadata is extracted by requesting data for the English language (and fallback to whatever is available). In the future, we'll want to unfurl by respecting the user's local device language. Some websites, like GoDaddy, are already localized based on the device's IP, but many aren't. - The website's description text should be limited by a certain number of characters, especially because it's outside our control. Exactly how much has not been decided yet, so it'll be done separately. - URL normalization can be tricky, so I implemented only the basics to help with caching. For example, the url https://status.im and HTTPS://status.im are considered identical. Also, a URL is considered valid for unfurling if its TLD exists according to publicsuffix.EffectiveTLDPlusOne. This was essential, otherwise the default Go url.Parse approach would consider many invalid URLs valid, and thus the server would waste resources trying to unfurl the unfurleable. ### Other requirements - If the message is edited, the link previews should reflect the edited text, not the original one. This has been aligned with the design team as well. - If the website's thumbnail or the favicon can't be fetched, just ignore them. The only mandatory piece of metadata is the website's title and URL. - Link previews in clients should be generated in near real-time, that is, as the user types, previews are updated. In mobile this performs very well, and it's what other clients like WhatsApp, Telegram, and Facebook do. ### Decisions - While the user typing in the input field, the client is constantly (debounced) asking status-go to parse the text and extract normalized URLs and then the client checks if they're already in its in-memory cache. If they are, no RPC call is made. I chose this approach to achieve the best possible performance in mobile and avoid the whole RPC overhead, since the chat experience is already not smooth enough. The mobile client uses URLs as cache keys in a hashmap, i.e. if the key is present, it means the preview is readily available (naive, but good enough for now). This decision also gave me more flexibility to find the best UX at this stage of the feature. - Due to the requirement that users should be able to see independent loading indicators for each link preview, when status-go can't unfurl a URL, it doesn't return it in the response. - As an initial implementation, I added the BLOB column unfurled_links to the user_messages table. The preview data is then serialized as JSON before being stored in this column. I felt that creating a separate table and the related code for this initial PR would be inconvenient. Is that reasonable to you? Once things stabilize I can create a proper table if we want to avoid this kind of solution with serialized columns.
2023-05-18 18:43:06 +00:00
func isValidLinkPreviewForProto(preview LinkPreview) bool {
return preview.Title != "" && preview.URL != "" &&
((preview.Thumbnail.DataURI == "" && preview.Thumbnail.Width == 0 && preview.Thumbnail.Height == 0) ||
(preview.Thumbnail.DataURI != "" && preview.Thumbnail.Width > 0 && preview.Thumbnail.Height > 0))
}
// ConvertLinkPreviewsToProto expects previews to be correctly sent by the
// client because we can't attempt to re-unfurl URLs at this point (it's
// actually undesirable). We run a basic validation as an additional safety net.
func (m *Message) ConvertLinkPreviewsToProto() ([]*protobuf.UnfurledLink, error) {
if len(m.LinkPreviews) == 0 {
return nil, nil
}
unfurledLinks := make([]*protobuf.UnfurledLink, 0, len(m.LinkPreviews))
for _, preview := range m.LinkPreviews {
// Do not process subsequent previews because we do expect all previews to
// be valid at this stage.
if !isValidLinkPreviewForProto(preview) {
return nil, fmt.Errorf("invalid link preview, url='%s'", preview.URL)
}
var payload []byte
var err error
if preview.Thumbnail.DataURI != "" {
payload, err = images.GetPayloadFromURI(preview.Thumbnail.DataURI)
if err != nil {
return nil, fmt.Errorf("could not get data URI payload, url='%s': %w", preview.URL, err)
}
}
ul := &protobuf.UnfurledLink{
Url: preview.URL,
Title: preview.Title,
Description: preview.Description,
ThumbnailWidth: uint32(preview.Thumbnail.Width),
ThumbnailHeight: uint32(preview.Thumbnail.Height),
ThumbnailPayload: payload,
}
unfurledLinks = append(unfurledLinks, ul)
}
return unfurledLinks, nil
}
func (m *Message) ConvertFromProtoToLinkPreviews(makeMediaServerURL func(msgID string, previewURL string) string) []LinkPreview {
var links []*protobuf.UnfurledLink
if links = m.GetUnfurledLinks(); links == nil {
return nil
}
previews := make([]LinkPreview, 0, len(links))
for _, link := range links {
parsedURL, err := url.Parse(link.Url)
var hostname string
// URL parsing in Go can fail with URLs that weren't correctly URL encoded.
// This shouldn't happen in general, but if an error happens we just reuse
// the full URL.
if err != nil {
hostname = link.Url
} else {
hostname = parsedURL.Hostname()
}
lp := LinkPreview{
Description: link.Description,
Hostname: hostname,
Title: link.Title,
URL: link.Url,
}
if payload := link.GetThumbnailPayload(); payload != nil {
lp.Thumbnail.Width = int(link.ThumbnailWidth)
lp.Thumbnail.Height = int(link.ThumbnailHeight)
lp.Thumbnail.URL = makeMediaServerURL(m.ID, link.Url)
}
previews = append(previews, lp)
}
return previews
}
func (m *Message) SetAlbumIDAndImagesCount(albumID string, imagesCount uint32) error {
imageMessage := m.GetImage()
if imageMessage == nil {
return errors.New("Image is empty")
}
imageMessage.AlbumId = albumID
imageMessage.AlbumImagesCount = imagesCount
m.Payload = &protobuf.ChatMessage_Image{Image: imageMessage}
return nil
}