466935 - TM: non-DEBUG JS_JIT_SPEW doesn't build, again

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Boris Zbarsky [:bzbarsky]]

It worked for a bit, but it's kinda broken now.  Main issues:

1)  Various JS_JIT_SPEW stuff in jstracer.cpp and jsregexp.cpp uses
    FramePCOffset, which is ifdef DEBUG.  Can we make it not DEBUG-only, or
    somehow condition it on JS_JIT_SPEW?
2)  jsregexp.cpp assumes that it can access LirBuffer::names ifdef JS_JIT_SPEW.
    But that's an LIR thing, not controlled by JS_JIT_SPEW at all.  It's
    controlled by the nanojit verbose_only macro.  In theory, that's defined
    when NJ_VERBOSE is defined, but just defining without defining DEBUG
    doesn't seem to build (more people making assumptions, I think).  Can we
    just wrap the LirBuffer::names access in DEBUG or NJ_VERBOSE ifdefs and not
    worry here?
3)  jsregexp uses VerboseWriter, which is also conditioned on NJ_VERBOSE.

I'm not sure whether jsregexp wants to just condition the whole thing on DEBUG instead of JS_JIT_SPEW and be done with it.

Comment 1

[Boris Zbarsky [:bzbarsky]]

For what it's worth, I'm happy to post some patches once brendan tells me how he wants FramePCOffset to work.

Comment 2

[Brendan Eich [:brendan]]

If FramePCOffset can be shown to be inlined only and to add no code size to optimized builds, please do remove the DEBUG ifdef. Thanks!

Need spin-off bug for items 2 and 3?

/be

Comment 3

[Boris Zbarsky [:bzbarsky]]

Attachment: Make it build again

This makes things compile.  I'm happy to spin off an additional bug for making the jsregexp logging work in opt builds if desired.  I did verify that there is no opt build codesize increase with this change.

Comment 4

[Boris Zbarsky [:bzbarsky]]

Pushed http://hg.mozilla.org/tracemonkey/rev/8aaf017a677c

Bug 466944 covers regexp stuff.