512887 - misc hydra fixes for callgraph work

Status: RESOLVED FIXED

(Developer Infrastructure :: Source Code Analysis, defect)

Description

[[e-mail]]

Attachment: patch v1

This rolls up a bunch of fixes.

* make hasDefaults property work (bug 510212)
* add resolve_path() utility function for resolving symlinks (to be removed later)
* add/tweak some macro stuff in gcc_compat.js
* fix type_string() to work with ptr-to-members and C++/C code in our tree, to order qualifiers (const etc) more consistently with g++, to work with vectors, and consistently throw if it gets confused - so we can spot errors easier. (I'm basically relying on this function being sane for serializing callgraph method types.)
* random fix to dehydra_attachClassStuff() to make it not die on classes with no bases
* tests to tickle most codepaths in type_string()

Comment 1

[[e-mail]]

Er, wrong component. ;)

Comment 2

[(dormant account)]

Comment on attachment 396947 [details] [diff] [review]
patch v1


>+function DECL_IS_BUILTIN(decl) {
>+  return DECL_SOURCE_LOCATION(decl).line == 0;
>+}

that's so not what tree.h does. These should be identical to gcc, otherwise it gets tricky to debug what's going wrong.

Seems ok otherwise, r+ with above fixed.

Comment 3

[[e-mail]]

Oh, does gcc/g++ #define USE_MAPPED_LOCATION?

Comment 4

[[e-mail]]

http://hg.mozilla.org/rewriting-and-analysis/dehydra/rev/6308679eb663

Comment 5

[(dormant account)]

http://hg.mozilla.org/rewriting-and-analysis/dehydra/rev/0485f8c977ea

resolve_path made a js string the wrong way