metro/docs/Bundling.md

61 lines
5.4 KiB
Markdown

---
id: bundling
title: Bundling
---
When bundling, each of the modules gets assigned a numeric id, meaning no dynamic requires are supported. Requires are changed by its numeric version, and modules are stored in different possible formats. Three different formats of bundling are supported:
## Plain bundle
This is the standard bundling format. In this format, all files are wrapped with a function call, then added to the global file. This is useful for environments that expect a JS only bundle (e.g. a browser). Just requiring the entry point with the `.bundle` extension should trigger a build of it.
## Indexed RAM bundle
This format composes the bundle as a binary file, which format has the following parts (all numbers are expressed in Little Endian):
* A magic number: a `uint32` must be located at the beginning of the file, with the value `0xFB0BD1E5`. This is used to verify the file.
* An offset table: the table is a sequence of `uint32` pairs, with a header
* For the header, two `uint32`s can be found: the length of the table, and the length of the startup code.
* For the pairs, they represent the offset in the file and the length of code module, in bytes.
* Each of the modules, finished by a null byte (`\0`).
```
` 0 1 2 3 4 5 6
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Magic number | Header size |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Startup code size | Module 0 offset |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Module 0 length | |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +
| |
+ ... +
| |
+ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| | Module n offset |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Module n length | Module 0 code | Module 0 code | ... | \0 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Module 1 code | Module 1 code | ... | \0 | |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ +
| |
+ ... +
| |
+ +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| | Module n code | Module n code | ... | \0 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+`
```
This structure is optimal for an environment that is able to load all code in memory at once:
* By using the offset table, one can load any module in constant time, where the code for module `x` is located at `file[(x + 3) * sizeof(uint32)]`. Since there is a null character (`\0`) separating all modules, usually length does not even need to be used, and the module can be loaded directly as an ASCIIZ string.
* Startup code is always found at `file[sizeof(uint32)]`.
This bundling is usually used by iOS.
## File RAM bundle
Each module is stored as a file, with the name `js-modules/${id}.js`, plus an extra file called `UNBUNDLE` is created, which its only content is the magic number, `0xFB0BD1E5`. Note that the `UNBUNDLE` file is created at the root.
This bundling is usually used by Android, since package contents are zipped, and access to a zipped file is much faster. If the indexed format was used instead, all the bundled should be unzipped at once to get the code for the corresponding module.