[cfe-commits] Fix search path for clang on latest DragonFly releases [revised patches]

John Marino draco at marino.st
Thu Nov 29 05:39:25 PST 2012


On 11/26/2012 14:38, Rafael EspĂ­ndola wrote:
>> However, what is the process to see this bug report 14417 accomplished? I
>> don't really want to stay subscribed to this mailing list for a single
>> patch.  I would have thought opening a bug report alone would have been
>> sufficient.
>
> No, we do code reviews on the list. I guess you could use phabricator
> to get messages just on this thread, but I never used it myself so I
> am not sure.
>
> This patch needs a test. Note that with this patch the driver will be
> passing multiple -Ls to the linker. It would be better to detect which
> gcc instalation will be used and pass just that one. See
> GCCInstallationDetector for example.
>

So I spent time over the last couple of days testing the patched Clang. 
  I found out quickly that I missed patching one file to change the 
header search patch, and also found that a GCC-built clang could not 
build clang (Problem with dwarf EH).

So I revised my patches and reposted to 
http://llvm.org/bugs/show_bug.cgi?id=14417 .  It passes the clang tests 
and a GCC-built clang can build clang now.

Some highlights:
* Only specify gcc 4.7 or gcc 4.4, not both
* removed rpath-link switches
* Added eh-frame-hdr support
* added gnu-hash support
* added dynamic export support
* Fixed crt* start and endfile switches, they were completely wrong due 
to last year's updates (mirroring FreeBSD's crt* changes)
* added pie support
* Added different libgcc spec for gcc 4.7 (differs greatly from 4.4)

It turns out that DragonFly already had a test, dragonfly.c, which 
passed with the previous patch.  However this new patch set required the 
dragonfly.c driver test to be updated as well.

The patches were written against clang 3.1:
http://llvm.org/bugs/attachment.cgi?id=9608&action=diff

If they are accepted, I'd like analogous patches to be applied to head 
and clang 3.2.  Do I need to generate those or are these 3.1 patches 
sufficient?

John





More information about the cfe-commits mailing list