[clang] Add new -fdebug-default-version flag.

[cmtice]

Summary

Want to decouple setting the DWARF version from enabling/disabling generation of debug info. This adds a new flag '-fdebug-default-version=N' that sets the version of DWARF to be generated, but does not turn on/off debug info generation. In particular this flag can be added to existing compilation commands to clarify which version of DWARF '-g' or '-gdwarf' will use, without overriding existing -gdwarf-N flags.

[dblaikie]

clang/lib/Driver/ToolChains/Clang.cpp

3246–3247

Looks like this should be on a single line to conform to LLVM convention (though might just be phabricator doing something weird)

If you can run clang-format over the change (not over the whole file) it should fix up issues like this. (there's various clang-format editor integrations - there's some google-internal documentation at go/clang-format that'll explain how to setup an auto-save hook that'll clang-format the changed lines so all your C++ code in the LLVM repository conforms to LLVM's coding conventions (well, those that can be expressed by clang-format))

3247–3248

Hmm, actually - why is this case necessary/what does it cover? I was hoping the case you added inside the "if (EmitDwarf)" case above would be all that was required (& the call to ParseDebugDefaultVersion would go inside there at the use, to reduce the variable scope/keep the code closer together).

6179–6181

Some clang-formatting required, but also could probably be written as:

if (DwarfVersion == 0)
  DwarfVersion = ParseDebugDefaultVersion(...);

if (DwarfVersion == 0)
  DwarfVersion = getToolChain().GetDefaultDwarfVersion();

To make these more symmetric?

[dblaikie]

clang/lib/Driver/ToolChains/Clang.cpp

3246–3247

Oh, there's also clang/tools/clang-format/git-clang-format for formatting anything in a git revision range.

[cmtice]

clang/lib/Driver/ToolChains/Clang.cpp

3246–3247

Ok, will do.

3247–3248

This covers the case where you are setting the default dwarf version without actually turning on/off debug info at all (there's no -gdwarf or -g option, so EmitDwarf is false).

6179–6181

Ok will do.

[MaskRay]

clang/test/CodeGen/debug-default-version.c

1 ↗

(On Diff #227768)

This looks like a clangDriver change. Maybe move the test to test/Driver ?

[MaskRay]

clang/test/CodeGen/debug-default-version.c

2 ↗

(On Diff #227768)

VER3 -> DWARF3 may be better, because you also check another debug info format codeview here.

[cmtice]

Made requested formatting changes.

[dblaikie]

clang/lib/Driver/ToolChains/Clang.cpp

3247–3248

Ah, perhaps I'm missing something - why should the DWARFVersion be set if EmitDwarf is false? No DWARF was requested, so the DWARFVersion probably shouldn't be set, I think?

[probinson]

Want to decouple setting the DWARF version from enabling/disabling generation of debug info.

Um, why?

[dblaikie]

In D69822#1733226, @probinson wrote:

Want to decouple setting the DWARF version from enabling/disabling generation of debug info.

Um, why?

If you've got a big build system with various users opting in/out of DWARF and you want to migrate to DWARFv5, say, but you can't add "-gdwarf-5" to your build system, because that'd turn on debug info in cases that don't need it - but it's easier to change the default than to modify all cases that enable dwarf across the codebase.

Open to discussion about whether this is a good/bad idea - but it'd be useful for Google at least & seemed low-cost enough to go this route.

[probinson]

clang/include/clang/Driver/Options.td

2002

Unexpected blank line here.

clang/lib/Driver/ToolChains/Clang.cpp

3237

precedence

clang/lib/Driver/ToolChains/CommonArgs.cpp

1151

Clang doesn't support DWARF v1. I expect nobody does at this point (DWARF 2 came out in 1993).

clang/test/CodeGen/debug-default-version.c

1 ↗

(On Diff #227768)

+1. It should be sufficient to use -### and check what gets passed as cc1 options.

24 ↗

(On Diff #227777)

Does that actually work? Last I checked, DWARF and COFF didn't play nicely, but I admit that was quite a while ago.

[aprantl]

Since this sounds like it is hidden inside of some other tooling — is passing the frontend option -dwarf-version=5 not an option?

[probinson]

In D69822#1733243, @dblaikie wrote:

In D69822#1733226, @probinson wrote:

Want to decouple setting the DWARF version from enabling/disabling generation of debug info.

Um, why?

If you've got a big build system with various users opting in/out of DWARF and you want to migrate to DWARFv5, say, but you can't add "-gdwarf-5" to your build system, because that'd turn on debug info in cases that don't need it - but it's easier to change the default than to modify all cases that enable dwarf across the codebase.

Open to discussion about whether this is a good/bad idea - but it'd be useful for Google at least & seemed low-cost enough to go this route.

Because you want the default to be based on your corporate environment rather than the target platform. The maze of twisty -g passages gets a new secret door. Oh well.

[dblaikie]

In D69822#1733258, @aprantl wrote:

Since this sounds like it is hidden inside of some other tooling — is passing the frontend option -dwarf-version=5 not an option?

"hidden inside of some other tooling" - in the same sense as most build flags are passed by a build system, yes. But generally we (Google specifically, all of us LLVM/Clang users in general) try not to use implementation details like that.

clang/test/CodeGen/debug-default-version.c

24 ↗

(On Diff #227777)

Existing test cases cover scenarios like this (see clang/test/CodeGen/dwarf-version.c), so this seems consistent to preserve that sort of functionality, however well it currently works.

[dblaikie]

In D69822#1733262, @probinson wrote:

In D69822#1733243, @dblaikie wrote:

In D69822#1733226, @probinson wrote:

Want to decouple setting the DWARF version from enabling/disabling generation of debug info.

Um, why?

If you've got a big build system with various users opting in/out of DWARF and you want to migrate to DWARFv5, say, but you can't add "-gdwarf-5" to your build system, because that'd turn on debug info in cases that don't need it - but it's easier to change the default than to modify all cases that enable dwarf across the codebase.

Open to discussion about whether this is a good/bad idea - but it'd be useful for Google at least & seemed low-cost enough to go this route.

Because you want the default to be based on your corporate environment rather than the target platform.

Sure - if we know our users have access to a modern debugger, but we don't own a whole platform. You could imagine someone shipping clang+debugger in a 3rd party/non-platform IDE might want something like this too/similar to the Google situation. We can't dictate what's suitable for all of Linux, but for the subset of users that are using a specific toolchain/situation, we can.

The maze of twisty -g passages gets a new secret door. Oh well.

If you find this to be a significant complication, I'd really like to discuss it further - I know there are some twisty things (one of the worst I created when it comes to -fsplit-dwarf-inlining, -gsplit-dwarf, -gmlt combinations) - but this seems pretty simple/tidy/orthogonal to me.

[probinson]

clang/include/clang/Driver/Options.td

1955

If this is specifically the default DWARF version, I think the word "dwarf" ought to be in the option name.

clang/lib/Driver/ToolChains/Clang.cpp

3148

I would rather see all the DWARF version weirdness all in one place (i.e., please move this down to the rest of it) particularly given that the "default" DWARF version now means two different things (something from the command line, and something based on the toolchain).

3249

After setting DWARFVersion above, when nothing was requested, this looks peculiar. Is DWARFVersion == 0 a proxy for -gcodeview ?

6173

I have to say, this sequence is a whole lot easier to follow than what's up in RenderDebugOptions. It would be nice if they were both so easy to understand.

Although it makes me wonder, does -fdebug-default-version go unclaimed here if there's a -gdwarf-2 on the command line?

[probinson]

In D69822#1733269, @dblaikie wrote:

In D69822#1733262, @probinson wrote:

The maze of twisty -g passages gets a new secret door. Oh well.

If you find this to be a significant complication, I'd really like to discuss it further - I know there are some twisty things (one of the worst I created when it comes to -fsplit-dwarf-inlining, -gsplit-dwarf, -gmlt combinations) - but this seems pretty simple/tidy/orthogonal to me.

No, no, if we can at least keep the twistiness in one place (or two I suppose, one for C/C++ and one for asm) it's not a real problem.

Maybe a strategic helper function could be useful to make those two places more obviously the same.

[dblaikie]

clang/lib/Driver/ToolChains/Clang.cpp

3249

Yep, DWARFVersion shouldn't be set when none was requested - see discussion above between myself & @cmtice - that should leave this code fine/correct.

6173

Once the "else if (DefaultDWARFVersion != 0)" clause is removed, and the ParseDebugDefaultVersion call is sunk down into the "if (EmitDwarf)" case - hopefully that'll simplify the RenderDebugOptions code enough to be reasonable?

I think the complication there is that "-g" can imply CodeView if nothing explicitly requests DWARF & the target is windows, whereas there's no support for that kind of CV fallback here? So some of that might be inherent.

& agreed - worth testing that -fdebug-default-version doesn't get a -Wunused warning if someone says "-fdebug-default-version 4 -gdwarf-2". @cmtice is that tested? You might be able to add -Werror to the existing test case's RUN lines to demonstrate that no such warnings are produced?

clang/lib/Driver/ToolChains/CommonArgs.cpp

1151

You'd recommend/request moving this bound up to 2 (to be consistent with the fact that clang supports -gdwarf-2 technically (even though we probably don't produce fully conformant DWARFv2 these days, I would guess))?

[probinson]

clang/lib/Driver/ToolChains/CommonArgs.cpp

1151

It wants to permit the exact same range as the -gdwarf-N options.

clang/test/CodeGen/debug-default-version.c

24 ↗

(On Diff #227777)

Ok.

[cmtice]

Made requested changes:

• renamed option to be dwarf-specific
• fixed spelling & blank line issues
• only set version if emit-dwarf is true
• move test to Driver directory

I *think* I got it all done...

[MaskRay]

In D69822#1734255, @cmtice wrote:

Made requested changes:

• renamed option to be dwarf-specific
• fixed spelling & blank line issues
• only set version if emit-dwarf is true
• move test to Driver directory

I *think* I got it all done...

It seems you uploaded the wrong diff, probably the original one...

[cmtice]

Try uploading correct diff file?

[cmtice]

Hmmm...I'm having upload issues.

clang/lib/Driver/ToolChains/Clang.cpp

3249

DWARFVersion is (will be) only set when the debug info kind is DWARF, i.e. when EmitDwarf is True. I apologize for the confusion -- David Blaikie and I has some misunderstandings about the original implementation intent. DWARFVersion == 0 ought only to result in codeview being generated/used if the user specified -g and codeview is the default debug format for the platform.

6173

The thought is that if the user actually chose a specific dwarf version, e.g. -gdwarf-2, then that value should override the general default, specified with -fdebug-default-version (soon to be
-fdwarf-default-version).

[cmtice]

re-try upload again.

[cmtice]

Ok, I think the upload was correct this time.

[dblaikie]

clang/include/clang/Driver/Options.td

1955

Can we haggle over this a bit?

My thinking behind -fdebug-default-version was consistency with other DWARF related flags:
-fdebug-compilation-dir
-fdebug-info-for-profiling
-fdebug-macro
-fdebug-types-section
-fdebug-ranges-base-address
-fdebug-prefix-map

We do have some -fdwarf:
-fdwarf2-cfi-asm
-fdwarf-directory-asm
-fdwarf-exceptions

So I'm personally inclined to sticking with -fdebug* as being all DWARF related/consistent there.

Thoughts?

clang/lib/Driver/ToolChains/Clang.cpp

3231

Can you move this down to the narrower scope:

if (unsigned DefaultDWARFVersion = ParseDefaultVersion(TC, Args))
  DWARFVersion = DefaultDWARFVersion;

(but maybe that runs into the -Wunused warning issues discussed in the other thread in this review)

6173

That's the desired behavior, yes (-gdwarf-2 overrides this new flag) - what @probinson is asking about is the behavior of -Wunused.

Clang's built-in flag handling does some work to warn on unused flags - it does this based on which flags are queried by the API.

The risk is that, because ParseDwarfDefaultVersion is only called under the "if (DwarfVersion == 0)" condition, which will be false if the user specified -gdwarf-2, for instance, then the ParseDwarfDefaultVersion won't be called & clang's unused flag handling may see the new flag as "unused" and warn about it.

Basically the question is (could you test this manually & confirm the behavior, and check that the automated tests cover this - which I think they can cover this with minimal effort by adding -Werror to some RUN lines in the tests), does this produce a warning:
clang -gdwarf-2 -fdebug-default-version=5 x.s

(& if it doesn't warn, it might be worth understanding why it doesn't, I guess, given the above hypothesis)

[cmtice]

dblaikie: The code, as written, parses the dwarf default version flag, whether or not DWARFVersion is 0, i.e. whether or not a -gdwarf-N flag was passed. It only USES the value if there's no overriding value.

[MaskRay]

clang/lib/Driver/ToolChains/Clang.cpp

3237

if (DefaultDWARFVersion) {
  // If the user specified a default DWARF version, that takes precedence
  // over the platform default.
  DWARFVersion = DefaultDWARFVersion;
} else {
  // Start with the platform default DWARF version.
  DWARFVersion = TC.GetDefaultDwarfVersion();
  assert(DWARFVersion && "toolchain default DWARF version must be nonzero");
}

6171

if (DwarfVersion == 0) {
  DwarfVersion = ParseDwarfDefaultVersion(getToolChain(), Args);
  if (DwarfVersion == 0)
    DwarfVersion = getToolChain().GetDefaultDwarfVersion();
}

This part constructs a -cc1as command line. It is not covered by a test. Can you add one?

[MaskRay]

clang/test/Driver/dwarf-default-version.c

1 ↗

(On Diff #227942)

-emit-llvm -> -###. @probinson mentioned that we just need to test -cc1 options.

We can use the following to check that codeview is not emitted.

// NOCODEVIEW-NOT: "-gcodeview"

[cmtice]

For the record, I double-checked and if we use the flag and don't check for it (i.e. if we move the parsing inside the EmitDwarf is True block) then -Werror does indeed complain about an unused command line argument.

[cmtice]

Make second call to ParseDwarfDefaultVersion unconditional (to match the first and avoid errors with -Werror).

[probinson]

The test is in the right place, now it needs to behave more like other driver tests. Sorry if it feels like I'm whaling on you, but the driver is a bit of a peculiar beast with an atypical testing mode. Taming it is harder than it looks.

clang/include/clang/Driver/Options.td

1955

Probably should have HelpText, maybe something like this:
"The default DWARF version to use, if a -g option causes DWARF debug info to be produced"
(it's probably not helpful to talk about overriding the target's default version, I think)

clang/lib/Driver/ToolChains/Clang.cpp

6172

This is the path for parsing assembler options (when feeding clang a .s file) and I believe as written, the combination -fdwarf-default-debug=2 -gdwarf-3 will yield an unused-option warning. So I'm pretty sure the test doesn't cover this path.
You can fake an assembler driver test in the current test file by using -x assembler (especially if you use the -### tactic I mention in the comments there), or you can add a second test file with a .s extension.

clang/test/Driver/dwarf-default-version.c

1 ↗

(On Diff #227942)

As a rule, driver tests should pass -### which causes the driver to print the command lines instead of running the commands. Then the checks would examine the command lines constructed by the driver, to verify they say what you want.
In this case, you'd look at the emitted -dwarf-version option and make sure it's what you want, and/or look for the option that says to do codeview.
The result is a test that is a more focused specifically on driver behavior, and also runs a bit faster (not starting any subprocesses etc).

10 ↗

(On Diff #227942)

indirectly

23 ↗

(On Diff #227942)

s/of/or/

[probinson]

clang/include/clang/Driver/Options.td

1955

If the HelpText mentions DWARF, I'm okay with an -fdebug prefix.

[probinson]

clang/lib/Driver/ToolChains/Clang.cpp

6171

Hi @MaskRay , your version would have the unclaimed-option problem.

[dblaikie]

clang/include/clang/Driver/Options.td

1955

Yep, I'd expect/be fine with all these options having something about DWARF in their long-form description.

@cmtice - would you mind switching this back to the original name & adding the relevant help text?

[cmtice]

Yes, I will change the flag name back to debug-default-version, and add help text mentioning dwarf. I'm in the process of trying to re-do the test cases as required (I'm a bit new to this so it's taking me a bit to figure this out).

[cmtice]

Rename flag to -fdebug-default-version; add help text, mentioning DWARF. Update tests to use -### and add assembler tests.

[dblaikie]

Looks good to me - but maybe give it a day to see if anyone else has further thoughts/that you've addressed their feedback too.

clang/test/Driver/debug-default-version.c

10–18

I'd probably skip these tests since overriding the platform default is already tested above (for the linux platform default) and I'm not sure there's much value add by testing that this overrides each platform default separately (given the platform default override is pretty orthogonal to which platform - it'd be pretty hard to introduce a bug that would break for one platform but not for all of them)

[MaskRay]

clang/test/Driver/debug-default-version.c

10–18

We may need -Werror in some or all of the tests.

[probinson]

Just a couple of typos, and I'm happy. I agree with the other reviewers on the last needed test tweaks.

clang/include/clang/Driver/Options.td

1955

HelpText does not want a final period.

clang/test/Driver/debug-default-version.c

11

indirectly

[MaskRay]

clang/test/Driver/debug-default-version.c

67

This is not correct. I suppose this to be "-dwarf-version=.

The pattern "-dwarf-version=" does not match "-dwarf-version=5"

[cmtice]

Remove period from help text. Fix tests.

[dblaikie]

clang/test/Driver/debug-default-version.c

38

Same issue as with the dwarf-version below.

It's probably easier to remove the "" (FileCheck arguments aren't quoted - these quotes are treated like any other character/as part of the match - so omitting them means FileCheck can match -debug-info-kind= no matter what comes before/after it, rather than specifically looking for a " immediately after the '='). (you could remove all the " if you like, I don't think they add a lot to the tests here)

[cmtice]

Fix NODEBUGINFO-NOT test.

[MaskRay]

LG after // NOCODEVIEW-NOT: "-gcodeview" and // NODWARF4-NOT: "-dwarf-version=4" are fixed.

[cmtice]

Remove quotes from -NOT tests.

[probinson]

Sigh, one last typo. I'm happy otherwise.

clang/test/Driver/debug-default-version.c

11

s/of you/or you/

[cmtice]

Fix small typo in comments.

clang/include/clang/Driver/Options.td

 def fdebug_ranges_base_address: Flag <["-"], "fdebug-ranges-base-address">, Group<f_Group>,
   Flags<[CC1Option]>, HelpText<"Use DWARF base address selection entries in debug_ranges">;
 def fno_debug_ranges_base_address: Flag <["-"], "fno-debug-ranges-base-address">, Group<f_Group>,
   Flags<[CC1Option]>;
 def fsplit_dwarf_inlining: Flag <["-"], "fsplit-dwarf-inlining">, Group<f_Group>,
   Flags<[CC1Option]>, HelpText<"Provide minimal debug info in the object/executable to facilitate online symbolication/stack traces in the absence of .dwo/.dwp files when using Split DWARF">;
 def fno_split_dwarf_inlining: Flag<["-"], "fno-split-dwarf-inlining">, Group<f_Group>,
   Flags<[CC1Option]>;
+def fdebug_default_version: Joined<["-"], "fdebug-default-version=">, Group<f_Group>, HelpText<"Default DWARF version to use, if a -g option caused DWARF debug info to be produced">;
 def fdebug_prefix_map_EQ
   : Joined<["-"], "fdebug-prefix-map=">, Group<f_Group>,
     Flags<[CC1Option,CC1AsOption]>,
     HelpText<"remap file source paths in debug info">;
 def fforce_dwarf_frame : Flag<["-"], "fforce-dwarf-frame">, Group<f_Group>, Flags<[CC1Option]>,
     HelpText<"Always emit a debug frame section">;
 def fno_force_dwarf_frame : Flag<["-"], "fno-force-dwarf-frame">, Group<f_Group>, Flags<[CC1Option]>,
     HelpText<"Don't always emit a debug frame section">;

 
 def gcodeview : Flag<["-"], "gcodeview">,
   HelpText<"Generate CodeView debug information">,
   Flags<[CC1Option, CC1AsOption, CoreOption]>;
 def gcodeview_ghash : Flag<["-"], "gcodeview-ghash">,
   HelpText<"Emit type record hashes in a .debug$H section">,
   Flags<[CC1Option, CoreOption]>;
 def gno_codeview_ghash : Flag<["-"], "gno-codeview-ghash">, Flags<[CoreOption]>;
 def ginline_line_tables : Flag<["-"], "ginline-line-tables">, Flags<[CoreOption]>;
 def gno_inline_line_tables : Flag<["-"], "gno-inline-line-tables">,
   Flags<[CC1Option, CoreOption]>, HelpText<"Don't emit inline line tables">;
 
 def gfull : Flag<["-"], "gfull">, Group<g_Group>;
 def gused : Flag<["-"], "gused">, Group<g_Group>;
 def gstabs : Joined<["-"], "gstabs">, Group<g_Group>, Flags<[Unsupported]>;
 def gcoff : Joined<["-"], "gcoff">, Group<g_Group>, Flags<[Unsupported]>;
 def gxcoff : Joined<["-"], "gxcoff">, Group<g_Group>, Flags<[Unsupported]>;

clang/lib/Driver/ToolChains/Clang.cpp

static void RenderDebugOptions(const ToolChain &TC, const Driver &D,
   // about what to pass from the driver to the frontend, but by the time they
   // reach cc1 they've been factored into three well-defined orthogonal choices:
   //  * what level of debug info to generate
   //  * what dwarf version to write
   //  * what debugger tuning to use
   // This avoids having to monkey around further in cc1 other than to disable
   // codeview if not running in a Windows environment. Perhaps even that
   // decision should be made in the driver as well though.
-  unsigned DWARFVersion = 0;
   llvm::DebuggerKind DebuggerTuning = TC.getDefaultDebuggerTuning();
 
   bool SplitDWARFInlining =
       Args.hasFlag(options::OPT_fsplit_dwarf_inlining,
                    options::OPT_fno_split_dwarf_inlining, true);
 
   Args.ClaimAllArgs(options::OPT_g_Group);
 
   Arg* SplitDWARFArg;
case codegenoptions::DIF_CodeView:
       EmitCodeView = true;
       break;
     case codegenoptions::DIF_DWARF:
       EmitDwarf = true;
       break;
     }
   }
 
+  unsigned DWARFVersion = 0;
+  unsigned DefaultDWARFVersion = ParseDebugDefaultVersion(TC, Args);
   if (EmitDwarf) {
     // Start with the platform default DWARF version
     DWARFVersion = TC.GetDefaultDwarfVersion();
     assert(DWARFVersion && "toolchain default DWARF version must be nonzero");
 
+    // If the user specified a default DWARF version, that takes precedence
+    // over the platform default.
+    if (DefaultDWARFVersion)
+      DWARFVersion = DefaultDWARFVersion;
+
     // Override with a user-specified DWARF version
     if (GDwarfN)
       if (auto ExplicitVersion = DwarfVersionNum(GDwarfN->getSpelling()))
         DWARFVersion = ExplicitVersion;
   }
 
   // -gline-directives-only supported only for the DWARF debug info.
   if (DWARFVersion == 0 && DebugInfoKind == codegenoptions::DebugDirectivesOnly)
     DebugInfoKind = codegenoptions::NoDebugInfo;
 
   // We ignore flag -gstrict-dwarf for now.
   // And we handle flag -grecord-gcc-switches later with DWARFDebugFlags.
   Args.ClaimAllArgs(options::OPT_g_flags_Group);
 
   // Column info is included by default for everything except SCE and
   // CodeView. Clang doesn't track end columns, just starting columns, which,
void ClangAs::ConstructJob(Compilation &C, const JobAction &JA,
   unsigned DwarfVersion = 0;
   Args.ClaimAllArgs(options::OPT_g_Group);
   if (Arg *A = Args.getLastArg(options::OPT_g_Group)) {
     WantDebug = !A->getOption().matches(options::OPT_g0) &&
                 !A->getOption().matches(options::OPT_ggdb0);
     if (WantDebug)
       DwarfVersion = DwarfVersionNum(A->getSpelling());
   }
+
+  unsigned DefaultDwarfVersion = ParseDebugDefaultVersion(getToolChain(),
+                                                          Args);
+  if (DwarfVersion == 0)
+    DwarfVersion = DefaultDwarfVersion;
+
   if (DwarfVersion == 0)
     DwarfVersion = getToolChain().GetDefaultDwarfVersion();
 
   codegenoptions::DebugInfoKind DebugInfoKind = codegenoptions::NoDebugInfo;
 
   if (SourceAction->getType() == types::TY_Asm ||
       SourceAction->getType() == types::TY_PP_Asm) {
     // You might think that it would be ok to set DebugInfoKind outside of
     // the guard for source type, however there is a test which asserts
     // that some assembler invocation receives no -debug-info-kind,
     // and it's not clear whether that test is just overly restrictive.
     DebugInfoKind = (WantDebug ? codegenoptions::LimitedDebugInfo
                                : codegenoptions::NoDebugInfo);
     // Add the -fdebug-compilation-dir flag if needed.

clang/lib/Driver/ToolChains/CommonArgs.h

void AddGoldPlugin(const ToolChain &ToolChain, const llvm::opt::ArgList &Args,
                    const InputInfo &Input, bool IsThinLTO);
 
 std::tuple<llvm::Reloc::Model, unsigned, bool>
 ParsePICArgs(const ToolChain &ToolChain, const llvm::opt::ArgList &Args);
 
 unsigned ParseFunctionAlignment(const ToolChain &TC,
                                 const llvm::opt::ArgList &Args);
 
+unsigned ParseDebugDefaultVersion(const ToolChain &TC,
+                                  const llvm::opt::ArgList &Args);
+
 void AddAssemblerKPIC(const ToolChain &ToolChain,
                       const llvm::opt::ArgList &Args,
                       llvm::opt::ArgStringList &CmdArgs);
 
 void addArchSpecificRPath(const ToolChain &TC, const llvm::opt::ArgList &Args,
                           llvm::opt::ArgStringList &CmdArgs);
 /// Returns true, if an OpenMP runtime has been added.
 bool addOpenMPRuntime(llvm::opt::ArgStringList &CmdArgs, const ToolChain &TC,

clang/lib/Driver/ToolChains/CommonArgs.cpp

unsigned tools::ParseFunctionAlignment(const ToolChain &TC,
 
   unsigned Value = 0;
   if (StringRef(A->getValue()).getAsInteger(10, Value) || Value > 65536)
     TC.getDriver().Diag(diag::err_drv_invalid_int_value)
         << A->getAsString(Args) << A->getValue();
   return Value ? llvm::Log2_32_Ceil(std::min(Value, 65536u)) : Value;
 }
 
+unsigned tools::ParseDebugDefaultVersion(const ToolChain &TC,
+                                         const ArgList &Args) {
+  const Arg *A = Args.getLastArg(options::OPT_fdebug_default_version);
+
+  if (!A)
+    return 0;
+
+  unsigned Value = 0;
+  if (StringRef(A->getValue()).getAsInteger(10, Value)
+      || Value > 5
+      || Value < 2)
+    TC.getDriver().Diag(diag::err_drv_invalid_int_value)
+        << A->getAsString(Args) << A->getValue();
+  return Value;
+}
+
 void tools::AddAssemblerKPIC(const ToolChain &ToolChain, const ArgList &Args,
                              ArgStringList &CmdArgs) {
   llvm::Reloc::Model RelocationModel;
   unsigned PICLevel;
   bool IsPIE;
   std::tie(RelocationModel, PICLevel, IsPIE) = ParsePICArgs(ToolChain, Args);
 
   if (RelocationModel != llvm::Reloc::Static)

clang/test/Driver/debug-default-version.c

+// RUN: %clang -### -Werror -target x86_64-linux-gnu -fdebug-default-version=4 -gdwarf-2 -S -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF2
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -gdwarf-3 -fdebug-default-version=4 -S -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF3
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -gdwarf-4 -fdebug-default-version=2 -S -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF4
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -gdwarf-5 -S -fdebug-default-version=2 -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF5
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -fdebug-default-version=5 -g -S -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF5
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -gdwarf -fdebug-default-version=2 -S -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF2
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -fdebug-default-version=4 -S -o - %s 2>&1 | FileCheck %s --check-prefixes=NODEBUGINFO,NODWARF4
+
+// Check which debug info formats we use on Windows. By default, in an MSVC
+// environment, we should use codeview. You can enable dwarf, which implicitly
+// disables codeview, of you can explicitly ask for both if you don't know how
+// the app will be debugged.
+// RUN: %clang -### -Werror -target i686-pc-windows-msvc -fdebug-default-version=2 -gdwarf -S -o - %s 2>&1 \
+// RUN:     | FileCheck %s --check-prefixes=DWARF2,NOCODEVIEW
+//     Explicitly request both.
+// RUN: %clang -### -Werror -target i686-pc-windows-msvc -fdebug-default-version=4 -gdwarf -gcodeview -S -o - %s 2>&1 \
+// RUN:     | FileCheck %s --check-prefixes=DWARF4,CODEVIEW
+
+
+// Do Assembler testing most of the same test cases as those above.
+
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -fdebug-default-version=5 -g -x assembler -c -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF5
+
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -fdebug-default-version=4 -gdwarf-2 -x assembler -c -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF2
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -gdwarf-3 -fdebug-default-version=4 -x assembler -c -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF3
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -gdwarf-4 -fdebug-default-version=2 -x assembler -c -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF4
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -gdwarf-5 -x assembler -c -fdebug-default-version=2 -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF5
+// RUN: %clang -### -Werror -target x86_64-linux-gnu -fdebug-default-version=5 -g -x assembler -c -o - %s 2>&1 | FileCheck %s --check-prefix=DWARF5
+
+int main (void) {
+  return 0;
+}
+
+// NOCODEVIEW-NOT: -gcodeview
+// CODEVIEW: "-gcodeview"
+
+// NODEBUGINFO-NOT: -debug-info-kind=
+
+// DWARF2: "-dwarf-version=2"
+// DWARF3: "-dwarf-version=3"
+// DWARF4: "-dwarf-version=4"
+// DWARF5: "-dwarf-version=5"
+
+// NOCODEVIEW-NOT: -gcodeview
+// NODWARF4-NOT: -dwarf-version=4