[PATCH] Handle properly somewhat special cases of -main-file-name

[Eric Christopher]

On Fri, Apr 4, 2014 at 9:08 AM, Lubos Lunak <l.lunak at centrum.cz> wrote:
> On Friday 04 of April 2014, David Blaikie wrote:
>> The comment in CGDebugInfo.cpp says that -main-file-name will contain
>> only the file name, with no path information. Are there cases where
>> that is not true when -main-file-name is passed from the Clang driver,
>> rather than by the user in your example below?
>>
>> If the driver provides this guarantee and the user violates it when
>> passing an undocumented flag manually, I don't see a need to support
>> it - but if there is such a need, it'd be good to
>> understand/documemnt/discuss it.
>
>  I think the driver always only passes the filename, but I do pass it
> explicitly, since I feed the source from stdin and the source or even its
> directory don't exist (distributed compiling, the same way with the patch
> for -dwarf-split-file).
>
>  And actually case 2) without using the option explicitly results in
> DW_AT_name becoming "/-" (I don't know why the driver doesn't simply pass the
> name as it is, which wouldn't need the code that attempts to rebuild it
> later).
>
>  As for "undocumented", the documentation status of the option is about the
> same like with many others - it's just in the .td file including its
> description. The description says quite clearly what the option does, it
> doesn't say it's internal, it works fine (except for these small problems), I
> even checked the sources to be sure, and I don't see why Clang should forbid
> usage of the option if I know what I'm doing (and would have to resort to
> ugly hacks otherwise).
>

There's no guarantee that the option will continue to exist at all so
I'd really prefer that you not use it. Any time you have to use
-Xclang to set something then it's not an exposed option that you
should be using.

That said, if we can come up with a structured set of uses for this
sort of compilation strategy rather than piecemeal then we should
discuss it and how it'll fit into the driver.

-eric


>> On Fri, Apr 4, 2014 at 5:02 AM, Lubos Lunak <l.lunak at centrum.cz> wrote:
>> >  The handling of -main-file-name in CGDebugInfo::CreateCompileUnit() can
>> > result in incorrect DW_AT_name in somewhat special cases:
>> >
>> > 1)
>> > $ touch /tmp/a.cpp
>> > $
>> > clang++ -Wall -c /tmp/a.cpp -g -o /tmp/a.o -Xclang -main-file-name
>> > -Xclang /new/path/a.cpp $ readelf -wi /tmp/a.o | grep DW_AT_name
>> >     <12>   DW_AT_name        : (indirect string, offset:
>> > 0x15): /tmp/new/path/a.cpp
>> >
>> > 2)
>> > $ touch /tmp/a.cpp
>> > $ cd /
>> > $ cat /tmp/a.cpp | clang++ -Wall -x
>> > c++ -c - -g -o /tmp/a.o -Xclang -main-file-name -Xclang a.cpp
>> > $ readelf -wi /tmp/a.o | grep DW_AT_name
>> >     <12>   DW_AT_name        : (indirect string, offset: 0x15): /a.cpp
>> >
>> >  The attached patch fixes those. Ok to commit?
> --
>  Lubos Lunak
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits