feat(wasm): WASM support (#301)

This implements WASM support. It uses a slightly modified protocol where a sender can
supply an `addr_mode` to be explicit about the addressing mode. `"abs"` is the default
absolute addressing mode, another option now is `rel:index` with the index of a module
in which case the instruction address is considered to be module relative. This change
also now lets existing users benefit of this by providing the relative address within any
other debug module as alternative.

CHANGELOG.md

@@ -10,6 +10,7 @@
 - Add `processing_pool_size` configuration option that allows to set the size of the processing pool. ([#273](https://github.com/getsentry/symbolicator/pull/273))
 - Use a dedicated `tmp` sub-directory in the cache directory to write temporary files into. ([#265](https://github.com/getsentry/symbolicator/pull/265))
 - Use STATSD_SERVER environment variable to set metrics.statsd configuration option ([#182](https://github.com/getsentry/symbolicator/pull/182))
+- Added WASM support. ([#301](https://github.com/getsentry/symbolicator/pull/301))
 
 ### Bug Fixes
 

docs/advanced/symbol-server-compatibility.md

@@ -75,6 +75,11 @@ Specifically, the code and debug identifiers are defined as follows:
   bidirectionally from the UUID.
 - **Debug ID:** The same UUID, amended by a `0` for age.
 
+**WASM**:
+
+- **Code ID:** The bytes as specified in the `build_id` custom section.
+- **Debug ID:** The same as code ID but truncated to 16 bytes + `0` for age.
+
 **PE** / **PDB**:
 
 - **Code ID:** The hex value of the `time_date_stamp` in the COFF header
@@ -266,6 +271,7 @@ The build-id hex representation is always provided in **lowercase**.
 
 - **ELF** (binary, potentially stripped)
 - **ELF** (debug info)
+- **WASM** (debug info)
 
 Symbol bundles are supported by adding a `.src.zip` prefix to the ELF:
 
@@ -297,7 +303,7 @@ The following layout types support this lookup:
 If you have no requirements to be compatible with another system you can also
 use the "unified" directory layout structure. This has the advantage that it's
 unified across all platforms and thus easier to manage. It can store breakpad
-files, PDBs, PEs and everything else.  The `symsorter` tool in the symbolicator
+files, PDBs, PEs and everything else. The `symsorter` tool in the symbolicator
 repository can automatically sort debug symbols into this format and also
 automatically create source bundles.
 
@@ -309,6 +315,7 @@ The debug id is in all cases lowercase in hex format and computed as follows:
 - **PDB**: `<Signature><Age>` (age in hex, not padded)
 - **ELF**: `<code_note_byte_sequence>`
 - **MachO**: `<uuid_bytes>`
+- **WASM**: `<BuildId>`
 
 The path format is then as follows:
 

docs/api/response.md

@@ -34,8 +34,8 @@ symbolication succeeds within a configured timeframe (around 20 seconds):
 
           // Frame information
           "instruction_addr": "0xfeedbeef",  // actual address of the frame
+          "addr_mode": "abs",                // address mode
           "sym_addr": "0xfeed0000",          // start address of the function
-          "line_addr": "0xfeedbe00",         // start address of the line
           "package": "/path/to/module.so",   // path to the module's code file
           "symbol": "__1cGmemset6FpviI_0_",  // original mangled function name
           "function": "memset",              // demangled short version of symbol
@@ -73,6 +73,15 @@ occurred during symbolication, such as missing symbol files or unresolvable
 addresses within symbols are reported as values for `status` in both modules and
 frames.
 
+## Note on Addresses
+
+Addresses (`instruction_addr` and `sym_addr`) can come in two versions. They
+can be absolute or relative. Symbolicator will always try to make addresses
+absolute but in some cases this cannot be done. For instance WASM modules do
+not have absolute addresses in which case the addresses stay relative. This is
+identified by the `addr_mode` property. When it's set to `"abs"` it means
+the addresses are absolute, when `"rel:X"` it's relative to module index `X`.
+
 ## Backoff Response
 
 If symbolication takes longer than the threshold `timeout`, the server instead

docs/api/symbolication.md

@@ -22,7 +22,8 @@ Content-Type: application/json
     {
       "frames": [
         {
-          "instruction_addr": "0xfeedbeef"
+          "instruction_addr": "0xfeedbeef",
+          "addr_mode": "rel:0"
         },
         ...
       ],
@@ -61,17 +62,18 @@ as well as external sources to pull symbols from:
 - `sources`: A list of descriptors for internal or external symbol sources. See
   [Sources](index.md).
 - `modules`: A list of code modules (aka debug images) that were loaded into the
-  process. All attributes other than `type` are required. The Symbolicator may
-  optimize lookups based on the `type` if present. Valid types are `macho`,
-  `pe`, `elf`. Invalid types are silently ignored. The Symbolicator still works
-  if the type is invalid, but less efficiently. However, a schematically valid
-  but _wrong_ type is fatal for finding symbols.
+  process. All attributes other than `type`, `image_addr` and `image_size` are
+  required. The Symbolicator may optimize lookups based on the `type` if present.
+  Valid types are `macho`, `pe`, `elf`. Invalid types are silently ignored. The
+  Symbolicator still works if the type is invalid, but less efficiently. However,
+  a schematically valid but _wrong_ type is fatal for finding symbols.
 - `threads`: A list of process threads to symbolicate.
-    - `registers`: Optional register values aiding symbolication heuristics. For
-      example, register values may be used to perform correction heuristics on the
-      instruction address of the top frame.
-    - `frames`: A list of frames with addresses. Arbitrary additional properties
-      may be passed with frames, but are discarded.
+  - `registers`: Optional register values aiding symbolication heuristics. For
+    example, register values may be used to perform correction heuristics on the
+    instruction address of the top frame.
+  - `frames`: A list of frames with addresses. Arbitrary additional properties
+    may be passed with frames, but are discarded. The `addr_mode` property
+    defines the beahvior of `instruction_addr`.
 
 ## Response