Summary:
Closes#82. This affords clients type-safety without needing to
verbosely annotate every node or edge passed into the graph functions.
It also enables graph algorithms to be more expressive in their types:
for instance, the merge function now clearly indicates from its type
that the first graph’s nodes are passed as the first argument to the
node reducer, and the second graph’s nodes to the second. Clients can
upgrade immediately by using `Graph<*, *>`.
Thankfully, Flow supports variance well enough for this all to be
possible without too much trouble.
Test Plan:
Existing unit tests pass statically and at runtime. I added a test case
to demonstrate that merging works covariantly.
To see some failures, change `string` to an incompatible type, like
`number`, in the definitions of `makeGraph` in test functions for
conservatively rejecting graphs with conflicting nodes/edges
(ll. 446, 462).
wchargin-branch: parameterize-graph
Summary:
Flow doesn’t allow us to specify variance annotations in generic
function parameters, and doesn’t allow coercing `Node<T>` to
`Node<mixed>`. This forces us to put `any`s in our code, which…works.
Paired with @dandelionmane.
Test Plan:
New unit tests trivially pass dynamically, and now pass statically
(failing before the changes to `graph.js`).
wchargin-branch: explicitly-typed-nodes-edges
Graph.addNode and Graph.addEdge now allow adding the same node or edge
multiple times, provided that the duplicate adds are trying to insert
identical content.
This came up while prototyping the GitHub plugin; rather than create
myriad subgraphs and merge them, I found it convenient to construct a
single graph and iteratively add nodes. Since the same node may be
discovered multiple times (most notably user identities), there was a
need for a "conservative add" abstraction that adds a node if it doesn't
exist yet, but errors only if multiple adds conflict.
Since this behavior is generic and highly conservative, it seemed
appropriate to include in the graph class itself.
Test Plan:
The unit tests have been updated to include the new behavior.
I moved sourcecred/tiny-example-repository to sourcecred/example-repo
as it's simpler to remember. I also unarchived it and added comments
to an issue, so that we can create a simple test for issue parsing.
This commit merely updates SourceCred to point to example-repo with
the regenerated canoncial output.
Summary:
It’s a whole new world of GraphQL! Our parser is now just a GraphQL
query that asks for exactly what we want and dumps it to a file. The
data exposed by the v4 API is also in a much nicer format than that of
the v3 API, so this is pretty much a universal improvement.
Currently, we do not handle pagination. We require that the repository
in question have fewer than a fixed number of issues, and comments per
issue, and reviews per PR, and review comments per PR, and so on. If
this limit is exceeded, the script will fail-fast with a nice error
message. To fix this, we’ll need to write a general-purpose pagination
API that allows traversing cursors at any level of the query.
Paired with @wchargin.
Test Plan:
Run
$ GITHUB_TOKEN="your_token_here" src/backend/fetchGitHubRepoTest.sh
and verify that it exits with 0. Note that if you change this script’s
repository from `tiny-example-repository` to `sourcecred`, the script
correctly fails and outputs a useful diff.
wchargin-branch: github-v4-graphql
* Factor evertide graph demo data to a new module
It would be helpful to make our standard tiny graph available to other
test and demo instances, outside of just graph.test.js. This way we can
use it as a test case for the Graph Explorer.
* Make App.js into skeleton for GraphExplorer
We make a very basic skeleton for the Graph Explorer as a basis
for future development.
This commit also removes the UserExplorer and FileExplorer from
App.js. Since we have changed the underlying data model, we are
unlikely to use the UserExplorer or FileExplorer in anything like
their current state, so they are effectively deprecated. I am deferring
removing them because it is nice to have some examples of working React
code to copy from, before the Graph Explorer is ready.
Test plan: run `yarn start`, and observe that the App displays the
words "Graph Explorer" underneath the "SourceCred Explorer" title bar.
Summary:
This commit adds `toJSON()` and `static fromJSON()` on `Graph`. The main
benefit at this time is that this gets us free interoperability with
Jest’s snapshot testing.
The implementation of `fromJSON` is not performance-tuned, and could
probably be significantly optimized.
See #65 for discussion.
Test Plan:
New unit tests added: `yarn flow && yarn test`.
wchargin-branch: make-graph-serializable
Summary:
This commit simplifies the implementation of `Graph` without changing
its interface. We now use the `AddressMap` for all four instance fields
of `Graph`.
Test Plan:
All existing tests pass, and coverage is maintained.
wchargin-branch: use-address-map-in-graph
Summary:
This commit reifies the concept of an `Addressable`, which is any object
that has a covariant `address: Address` attribute, and implements a
simple data structure for storing addressable items keyed against their
addresses. Instances of `AddressMap` can replace the four fields of
`Graph`:
```js
_nodes: AddressMap<Node<mixed>>;
_edges: AddressMap<Edge<mixed>>;
_outEdges: AddressMap<{|+address: Address, +edges: Address[]|}>;
_inEdges: AddressMap<{|+address: Address, +edges: Address[]|}>;
```
Test Plan:
New unit tests included, with 100% coverage: `yarn flow && yarn test`.
wchargin-branch: address-map
Summary:
We’re stripping down the payload types for the GitHub plugin, to only
include what we expect to use immediately. In doing so, we take the
opportunity to make the typing a little stronger, so that we can ensure
that the `type` field of a specific type of payload is set to a
particular constant.
Paired with @dandelionmane.
Test Plan:
Adding these lines to `githubPlugin.js` and running `yarn flow`
indicates that the typechecking is working as expected:
```js
("ISSUE" : NodeType); // works
("WEIRD" : NodeType); // fails
("AUTHORSHIP" : EdgeType); // works
("UNEXPECTED" : EdgeType); // fails
```
wchargin-branch: github-plugin-payload-types
Summary:
`graph.js` coverage is now 100% :-)
Test Plan:
`yarn jest --env=jsdom --coverage` shows no uncovered lines for
`graph.js`, and no failing tests.
wchargin-branch: coverage-gremlin
Summary:
Merging graphs will be a common operation. At a per-plugin level, it
will often be useful to build up graphs by creating many very small
graphs and then merging them together. At a cross-project level, we will
need to merge graphs across repositories to gain an understanding of how
value flows among these repositories. It’s important that the core graph
type provide useful functions for merging; this commit adds them.
Test Plan:
New unit tests added; run `yarn flow && yarn test`.
wchargin-branch: graph-merge
Summary:
We need this for testing graph equality: deep-equality is not sufficient
because two graphs can be logically equal even if, say, two nodes are
added in different orders.
This commit adds a dependency on `lodash.isequal` for deep equality.
Test Plan:
New unit tests added. Run `yarn flow && yarn test`.
wchargin-branch: graph-equals
Summary:
In merging #54, there was a semantic merge conflict that was not also a
textual merge conflict; this created a failure that only appeared once
that commit was merged.
We propose that to fix this in the future, we only merge commits that
are directly ahead of master.
Test Plan:
This fixes `yarn flow` and `yarn test`.
wchargin-branch: fix-merge-conflict
Summary:
Again: we assume these invariants, so we may as well encode them.
We should just keep in mind that non-Flow users may wantonly violate
these, so we should still code defensively.
wchargin-branch: readonly-exact
Summary:
These will make nicer error functions in cases where static analysis
doesn’t detect the pollution: e.g., a user isn’t using Flow, or an
expression like `arr[0]` introduces an `undefined`.
Paired with @dandelionmane.
Test Plan:
New unit tests added. Run `yarn test`.
wchargin-branch: null-undefined-check
Create an 'advancedMealGraph' test case
The advancedMealGraph will be a grab-all that holds all advanced and
edge behaviors, e.g. the crab-self-referential loop, and the case
where there are multiple directed edges between the same two nodes.
Aggregating them into one test case will make it easier to test more
complex behaviors, like graph merging and serialization, on the
edge case graphs. However, it's still nice to have the simple graph
so that we can test simple things too. The specific tests for edge
case behavior are left mostly unchanged, in that they start from the
simple graph and add just the advanced feature that they want to test.
Summary:
Without these functions, it is not possible to meaningfully operate on
an arbitrary graph.
Paired with @dandelionmane.
Test Plan:
New unit tests included. Run `yarn flow && yarn test`.
wchargin-branch: get-all
Summary:
We’ve realized that `u: Edge<T>` implies `u: Node<T>`. That certainly
wasn’t what we were expecting! We might want something like that
eventually, to capture the fact that valuations are themselves valuable,
but for now the type system should encode the assumptions that we’re
actually making. See also #50.
Paired with @dandelionmane.
wchargin-branch: exact-types
Summary:
We had planned to expose our core types as simple Plain Old JavaScript
Objects, with accompanying standalone functions to act directly on these
data structures. We chose this instead of creating `class`es for the
types because it simplifies serialization interop: it obviates the need
for serialization and deserialization functions, because the code is
separated from the data entirely. Reconsidering, we now think that the
convenience benefits of using classes probably outweigh these
serialization cons. Furthermore, this design enables us to separate
ancillary data structures and caches from the raw data, presenting a
cleaner API for consumers of the data.
This commit introduces a `Graph` class and some related logic. With lots
of tests! And 100% code coverage! :-)
Paired with @dandelionmane.
Test Plan:
Run `yarn flow && yarn test` to see the new tests.
wchargin-branch: graph-class
Summary:
The main problem with having these fields on the node is that this
presents the illusion that the API surface area is larger than it
actually is. Clients with reference to a node object could
somewhat-reasonably expect that mutating these fields would be
sufficient to update the structure of the graph, but this isn’t the case
(as the edge objects would need to be updated, too). It’s a nice
semantic bonus, too, as edges aren’t conceptually “part of” nodes.
wchargin-branch: top-level-edges
Summary:
This is an experiment. There are a couple diffferent meanings of
“weight” in play: most prominently, weights assigned by plugins versus
those suitable for comparison among other arbitrary weights. We’re not
sure what the right thing is to put in the actual graph object, so we’re
going to think about this a bit more before adding the field back in.
wchargin-branch: remove-weights
Summary:
The “ID” parts were left-over from the Great Address Migration, and we
think that abbreviations are fine here, anyway.
Test Plan:
`yarn flow && yarn test`
wchargin-branch: src-dst-rename
design.md and overview.md both describe a vision of sourcecred
in which it is a measure of credit, and an explicit cryptographic
token. right now, SourceCred is more focused on just measuring credit,
with the expectation that cryptoincentives can be added on later.
Removing these outdated documents will reduce confusion; they may
be re-written and re-added later.
Summary:
The sourcecred/tiny-example-repository repository stores some example
data that we can use to generate test cases. As of now, the repository
has been archived so that its state is stable. This commit checks in the
result of our scraper on the repository.
wchargin-branch: example-data
Reorganize the code so that we have a single package.json file, which is at the root.
All source code now lives under `src`, separated into `src/backend` and `src/explorer`.
Test plan:
- run `yarn start` - it works
- run `yarn test` - it finds the tests (all in src/explorer) and they pass
- run `yarn flow` - it works. (tested with an error, that works too)
- run `yarn prettify` - it finds all the js files and writes to them
Summary:
This tool grabs all the information that we think will be relevant for a
first-pass implementation of the SourceCred project graph. It includes a
tool to save the results to disk so that we avoid needlessly hitting the
GitHub API over and over.
Paired with @dandelionmane.
Test Plan:
The API doesn’t have tests, because we didn’t think that they would
provide much marginal value. But here’s how you invoke it:
node bin/fetchAndPrintGitHubRepo.js sourcecred sourcecred "${TOKEN}" >/tmp/out
to crawl the repository `sourcecred/sourcecred` with the given API
token.
wchargin-branch: grab-github-data
This reflects our current vision that SourceCred is aimed at valuing open-source contributions via a credit graph, not on directly creating cryptotokens
* Use font roboto
* Change title.
* Add grid gap
* Restyle app; blue header, and css->inline.
* Further visual cleanup.
- Apply consistent formatting to the two plugin panes.
- Add scroll to the User Explorer.
- Both have white backgrounds with centered titles.
Setup prettier precommit, with yarn prettify to reformat.
- Prettier runs as a precommit hook and fails if code is improperly formatted.
- Run `yarn prettify` to reformat the project.
Summary:
Closes#17; see discussion there.
This commit uses the `babel-plugin-flow-react-proptypes` package to
automatically create PropType definitions from components that are typed
with Flow. It simultaneously updates all of our existing components to
be typed with Flow. As a result, we have both static and dynamic type
checking.
Test Plan:
Note that `yarn test` and `yarn flow` report no errors, and that there
are no prop validation errors at runtime with `yarn start`.
Then, apply the following patch:
```diff
diff --git a/explorer/src/UserExplorer.js b/explorer/src/UserExplorer.js
index bb574cd..636a10d 100644
--- a/explorer/src/UserExplorer.js
+++ b/explorer/src/UserExplorer.js
@@ -18,7 +18,7 @@ export class UserExplorer extends Component<{
.sort((a,b) => b[1] - a[1]);
const entries = sortedUserWeightTuples.map(authorWeight => {
const [author, weight] = authorWeight;
- return <UserEntry userId={author} weight={weight} key={author}/>
+ return <UserEntry userId={55} weight={weight} key={author}/>
});
return <div className="user-explorer">
<h3> User Explorer </h3>
```
Note that `yarn test` fails (the `App.test.js` E2E rendering test),
`yarn flow` fails, and there is a runtime prop validation error.
wchargin-branch: autogenerate-proptypes
Summary:
This commit was created by performing the following steps:
- `rm -r node_modules`
- `yarn install`
- `yarn eject`
- `rm -r node_modules`
- `yarn install`
on an environment wit Node v6.11.1.
The final `rm -r node_modules && yarn install` ensures that the
`yarn.lock` file is updated accordingly.
Test Plan:
`yarn test`, `yarn flow`, and `yarn start` all still do the right thing.
wchargin-branch: eject
Summary:
This should ensure that there aren’t any runtime PropTypes errors.
The change to reimplement `Object.entries` is due to the fact that my
local Node environment (v6.11.1) does not yet support that ESnext
function.
Test Plan:
Apply the following diff (`git apply`):
```diff
diff --git a/explorer/src/UserExplorer.js b/explorer/src/UserExplorer.js
index 8da252d..cdc1370 100644
--- a/explorer/src/UserExplorer.js
+++ b/explorer/src/UserExplorer.js
@@ -10,7 +10,7 @@ export class UserExplorer extends Component {
static propTypes = {
selectedPath: PropTypes.string.isRequired,
selectedUser: PropTypes.string,
- onSelectUser: PropTypes.func.isRequired,
+ onSelectUser: PropTypes.number.isRequired,
data: commitUtilsPropTypes.commitData.isRequired,
}
```
Then, `yarn test` should fail. Revert the diff, and `yarn test` should
pass.
wchargin-branch: react-proptypes-test
William Chargin
Dandelion Mané