Summary: | Remove alignment code from IPC coders | ||||||
---|---|---|---|---|---|---|---|
Product: | WebKit | Reporter: | Antti Koivisto <koivisto> | ||||
Component: | WebKit2 | Assignee: | Nobody <webkit-unassigned> | ||||
Status: | REOPENED --- | ||||||
Severity: | Normal | CC: | ahmad.saleem792, andersca, ap, cdumez, commit-queue, kkinnunen, ossy | ||||
Priority: | P2 | ||||||
Version: | 528+ (Nightly build) | ||||||
Hardware: | Unspecified | ||||||
OS: | Unspecified | ||||||
Bug Depends on: | 139081, 139239 | ||||||
Bug Blocks: | |||||||
Attachments: |
|
Description
Antti Koivisto
2014-11-21 05:40:13 PST
Created attachment 242048 [details]
patch
Re-opened since this is blocked by bug 139239 If you are going to reland this patch once, please integrate the buildfix from bug139081. Thanks. Is there any plan to land it again with the iOS fix and the fix in bug139081? I think Anders said he wanted to do something different here. (In reply to comment #5) > Is there any plan to land it again with the iOS fix and the fix in bug139081? (In reply to comment #6) > I think Anders said he wanted to do something different here. Anders? ping? ping? This r+ patch landed: https://github.com/WebKit/WebKit/commit/395664aba4d8e744cf9e206e326259ae171d8bd0 but got backed out: https://github.com/WebKit/WebKit/commit/92c313f6d581fe6d372217e388c4ee4e9f1cc80d Is this needed anymore? Thanks! I'm not sure at 2014, but in 2022 I think the alignment is used to ensure that the possible in-place references of data are correctly aligned. E.g. mostly the case where we would refer to Span<float> that exist in the stream IPC buffer or mach message buffer, and we'd prefer not to copy it (say it's 50mbs of floats for WebGL). There might be cases where the alignment could be omitted. There might be call sequences where the invariants are broken so that the above expectation doesn't actually hold, we don't have the tests yet for this. |