1550616 - Touching TokenStream.h forces far, far too many files to be recompiled

Status: RESOLVED FIXED

(Core :: JavaScript Engine, task)

Description

[Jeff Walden [:Waldo]]

A number of headers directly or indirectly #include "frontend/TokenStream.h" when they need none or almost none of it. By carefully modifying a few of those headers, it's possible to significantly reduce the number of *.cpp files that require recompilation.

I've been touching TokenStream.h enough lately, for small changes, that this is seriously biting me. Let's reduce the dependency graph a bit.

Comment 1

[Jeff Walden [:Waldo]]

Attachment: Bug 1550616 - Move GeneralTokenStreamChars::setFunctionStart to GeneralParser::setFunctionStartAtCurrentToken in Parser.cpp, eliminating one reason SharedContext.h depends on TokenStream.h. r=tcampbell

Comment 2

[Jeff Walden [:Waldo]]

Attachment: Bug 1550616 - Make FunctionBox::setEnd take the actual uint32_t end, not TokenStreamAnyChars&, and add ParserBase::setFunctionEndFromCurrentToken that calls it passing the correct value, to eliminate another SharedContext.h dependency on...

...TokenStream.h. r=tcampbell

Depends on D30574

Comment 3

[Jeff Walden [:Waldo]]

Attachment: Bug 1550616 - Make SharedContext.h not #include TokenStream.h now that it doesn't depend on it. r=tcampbell

Depends on D30575

Comment 4

[Jeff Walden [:Waldo]]

Attachment: Bug 1550616 - Move Token-related types to a new frontend/Token.h header so needing something like TokenPos doesn't demand a dependency on TokenStream.h. (ParseNode.h is a notable instance of this.) r=tcampbell

Depends on D30576

Comment 5

[Jeff Walden [:Waldo]]

Attachment: Bug 1550616 - Move ModuleBuilder into its own header so that module object-related data structures don't depend on EitherParser.h (and therefore all of parsing and tokenizing). r=tcampbell

Depends on D30577

Comment 6

[Jeff Walden [:Waldo]]

Attachment: Bug 1550616 - Don't #include EitherParser.h in BytecodeCompilation.h when a simple forward-declaration will do. r=tcampbell

Depends on D30578

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/dc57399bbbe3
Move GeneralTokenStreamChars::setFunctionStart to GeneralParser::setFunctionStartAtCurrentToken in Parser.cpp, eliminating one reason SharedContext.h depends on TokenStream.h.  r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/59f2744f50c5
Make FunctionBox::setEnd take the actual uint32_t end, not TokenStreamAnyChars&, and add ParserBase::setFunctionEndFromCurrentToken that calls it passing the correct value, to eliminate another SharedContext.h dependency on... r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/5fd47cd91b82
Make SharedContext.h not #include TokenStream.h now that it doesn't depend on it.  r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/1b35028d64e4
Move Token-related types to a new frontend/Token.h header so needing something like TokenPos doesn't demand a dependency on TokenStream.h.  (ParseNode.h is a notable instance of this.)  r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/b435e9bb58cc
Move ModuleBuilder into its own header so that module object-related data structures don't depend on EitherParser.h (and therefore all of parsing and tokenizing).  r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/f1a3220fc10b
Don't #include EitherParser.h in BytecodeCompilation.h when a simple forward-declaration will do.  r=tcampbell

Comment 8

[Noemi Erli[:noemi_erli]]

Backed out 6 changesets (Bug 1550616) for causing build bustages in MacroAssembler-arm64.h

Push with failures: https://treeherder.mozilla.org/#/jobs?repo=autoland&resultStatus=testfailed%2Cbusted%2Cexception&revision=f1a3220fc10b57240e57bde4bfeba9745013dffe

Failure log: https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=245998300&repo=autoland&lineNumber=6812

Backout: https://hg.mozilla.org/integration/autoland/rev/464bde42a1588d89f7822d175f86db7f93d211ea

Comment 9

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/703783d5484d
Move GeneralTokenStreamChars::setFunctionStart to GeneralParser::setFunctionStartAtCurrentToken in Parser.cpp, eliminating one reason SharedContext.h depends on TokenStream.h.  r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/c706eab3c639
Make FunctionBox::setEnd take the actual uint32_t end, not TokenStreamAnyChars&, and add ParserBase::setFunctionEndFromCurrentToken that calls it passing the correct value, to eliminate another SharedContext.h dependency on... r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/78749757be96
Make SharedContext.h not #include TokenStream.h now that it doesn't depend on it.  r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/0e910372a956
Move Token-related types to a new frontend/Token.h header so needing something like TokenPos doesn't demand a dependency on TokenStream.h.  (ParseNode.h is a notable instance of this.)  r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/0fde18f5d16f
Move ModuleBuilder into its own header so that module object-related data structures don't depend on EitherParser.h (and therefore all of parsing and tokenizing).  r=tcampbell
https://hg.mozilla.org/integration/autoland/rev/cb00aa26b348
Don't #include EitherParser.h in BytecodeCompilation.h when a simple forward-declaration will do.  r=tcampbell

Comment 10

[Alexandru Michis [:malexandru]]

https://hg.mozilla.org/mozilla-central/rev/703783d5484d
https://hg.mozilla.org/mozilla-central/rev/c706eab3c639
https://hg.mozilla.org/mozilla-central/rev/78749757be96
https://hg.mozilla.org/mozilla-central/rev/0e910372a956
https://hg.mozilla.org/mozilla-central/rev/0fde18f5d16f
https://hg.mozilla.org/mozilla-central/rev/cb00aa26b348