Support arbitrary -gX debug options, and limit DWARF to line-tables-only

[kripken]

Fixes #9611

Sadly this adds some complexity, but I tried hard to keep
it all in one area, and not spread around.

[dschuff]

After looking at dwarf-splitting, I think we're going to want to make at least the compile-part of it a fully general passthrough, because the flag for splitting is -gsplit-dwarf[=single]

[dschuff]

This is also weird because g1 is more debug info than line-tables-only but we are modeling it as level 3.
Also we may want the ability to control generation of source maps independently from names sections and dwarf sections.
Maybe we need to make a complete separation between flags used at compile time and link time, where we pass through at compile times and accept a different set of flags at link time. It's probably already the case that users are doing special things for link anyway because build systems don't pass -g at link time by default.

[kripken]

Added arbitrary passthrough support, and made it limit debug info to what we can do for now, which is line tables.

[dschuff]

Doesn't this mean that you can never get g1-level debug info from the frontend?

[kripken]

Oh, does -g1 mean something useful for clang? clang --help doesn't show anything.

[dschuff]

Clang's options are not well documented. But the behavior tends to match https://gcc.gnu.org/onlinedocs/gcc-7.5.0/gcc/Debugging-Options.html#Debugging-Opti

…

On Wed, Nov 20, 2019 at 12:51 PM Alon Zakai ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In emcc.py <#9872 (comment)> : > options.requested_debug = newargs[i] - newargs[i] = '' + requested_level = newargs[i][2:] or '3' + if is_int(requested_level): + # the -gX value is the debug level (-g1, -g2, etc.) + options.debug_level = validate_arg_level(requested_level, 4, 'Invalid debug level: ' + newargs[i]) + # if we don't need to preserve LLVM debug info, do not keep this flag + # for clang + if options.debug_level < 3: Oh, does -g1 mean something useful for clang? clang --help doesn't show anything. — You are receiving this because your review was requested. Reply to this email directly, view it on GitHub <#9872?email_source=notifications&email_token=AAISEKGELJGCRKXRKYYEG2LQUWPLTA5CNFSM4JPLMMU2YY3PNVWWK3TUL52HS4DFWFIHK3DMKJSXC5LFON2FEZLWNFSXPKTDN5WW2ZLOORPWSZGOCMLNVKI#discussion_r348738103>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAISEKH45W4PHQPHP65PDXTQUWPLTANCNFSM4JPLMMUQ> .

[kripken]

I see. Then yes, -g1 in emcc is different from gcc/clang, and we've not supported the gcc/clang meaning. This PR doesn't change that, but yeah, that's unfortunate.

[dschuff]

Anyway this is a functionality improvement for now without breaking the current model, although I think it does illustrate why I think we want to maybe split compile and link-time debug options more carefully. I think most of the interesting and/or weird and/or not-applicable-off-the-web functionality is done at link time.

[kripken]

True. Maybe that refactoring could be helped by @sbc100's planned refactoring of an emlink type tool, which separates the linking part completely.