[cfe-dev] GDB test suite & debug info validation

[Eric Christopher]

On Jul 30, 2012, at 10:59 AM, David Blaikie <dblaikie at gmail.com> wrote:

> Hi guys,
> 
> [mostly directed at Eric & Nick who I've talked to a bit about this
> before & seem to be the ones most involved in this aspect of LLVM &
> Clang (I've only included cfe-dev at the moment, if anyone feels
> this'd be more appropriate spread over llvm-dev too, or instead, or
> just taken off-list with relevant parties, let me know)]
> 
> I'm interested in getting some kind of continuous debug info
> validation going as well as something locally-runnable for push-button
> execution when iterating locally.

Yay!

> 
> Eric - you mentioned you guys had some amount of gdb test suite
> execution going on? Could you point me to any upstream/public
> artifacts of that implementation. I got the impression this was a
> buildbot sort of situation & not a convenient locally
> runnable/iterative setup? Also getting some more recent (>4.2) suites
> running somewhere was mentioned - I figure once I can repro your
> current setup I can see about substituting in other suites.
> 

The testsuite is located here:

https://$USER@llvm.org/svn/llvm-project/clang-tests/trunk

and works via 'make' in the gdb-1472-testsuite directory with this patch:

Index: gdb-1472-testsuite/Makefile
===================================================================
--- gdb-1472-testsuite/Makefile	(revision 160058)
+++ gdb-1472-testsuite/Makefile	(working copy)
@@ -5,8 +5,8 @@
 objdir := $(shell pwd)/obj
 
 CCC_ADD_ARGS := -Qunused-arguments,-std-default=gnu89,-fblocks,-Wno-return-type,-Wno-unused-value,-Wno-switch-enum,-Wno-implicit-int,-Wno-incompatible-pointer-types,-Wno-switch
-CC_UNDER_TEST := /Developer/usr/bin/clang
-CXX_UNDER_TEST := /Developer/usr/bin/clang++
+CC_UNDER_TEST := clang
+CXX_UNDER_TEST := clang++
 CC_UNDER_TEST_FLAGS :=
 CXX_UNDER_TEST_FLAGS :=

(otherwise it obviously runs against the installed clang)

there are a bunch of known failures (see the FAIL.txt) file under expected results. Not all of these are "expected to fail and not our fault", but they're "expected" at the moment - depending on host OS.

It's just also run by the buildbot and the buildbot compares against that file to determine whether or not we have something new failing at the bot level. But yes, a newer gdb and testsuite is among my fondest desires here - at least knowing if something is failing for someone else would be good.

-eric