* Re: Changing the C/C++ compiler for gdb testsuite runs
@ 2002-01-10 8:58 Michael Elizabeth Chastain
2002-01-10 11:10 ` Kevin Buettner
0 siblings, 1 reply; 12+ messages in thread
From: Michael Elizabeth Chastain @ 2002-01-10 8:58 UTC (permalink / raw)
To: Richard.Earnshaw; +Cc: gdb
Richard Earnshaw writes:
> Is there a way of forcing the GDB testsuite to use a particular compiler
> that hasn't yet been installed (I have a gcc-3 compiler from a build tree
> that I'd like to use to run the tests, but I don't particularly want to
> have to install it first).
I build my compilers with "prefix=/blah/blah/2002-01-10/blah ...."
and then I go ahead and install the compiler. When I run the gdb test
suite, I set $PATH to the place where the compiler-under-test lives.
You may need to set $LD_LIBRARY_PATH as well if you build gcc with shared
libraries (libgcc and libstdc++).
If you build gcc and gdb in a unified build tree then dejagnu will find
and use the uninstalled compiler in the unified build tree, even in
preference to $PATH. That behaviour actually got in my way so I stopped
using unified trees.
Michael C
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Changing the C/C++ compiler for gdb testsuite runs
2002-01-10 8:58 Changing the C/C++ compiler for gdb testsuite runs Michael Elizabeth Chastain
@ 2002-01-10 11:10 ` Kevin Buettner
2002-01-10 11:30 ` Daniel Jacobowitz
0 siblings, 1 reply; 12+ messages in thread
From: Kevin Buettner @ 2002-01-10 11:10 UTC (permalink / raw)
To: Michael Elizabeth Chastain, Richard.Earnshaw; +Cc: gdb
On Jan 10, 10:58am, Michael Elizabeth Chastain wrote:
> If you build gcc and gdb in a unified build tree then dejagnu will find
> and use the uninstalled compiler in the unified build tree, even in
> preference to $PATH. That behaviour actually got in my way so I stopped
> using unified trees.
Here's one way to use a different compiler in a unified tree...
1) In gdb/testsuite (in your build directory), do
make site.exp
2) Add lines similar to the following to the end of
gdb/testsuite/site.exp:
set CC_FOR_TARGET "/some/other/path/to/gcc"
set CXX_FOR_TARGET "/some/other/path/to/g++"
If anyone knows of other (more elegant) tricks that can be used to
accomplish the same thing, I'd like to hear about them...
Kevin
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Changing the C/C++ compiler for gdb testsuite runs
2002-01-10 11:10 ` Kevin Buettner
@ 2002-01-10 11:30 ` Daniel Jacobowitz
2002-01-15 10:24 ` Daniel Jacobowitz
0 siblings, 1 reply; 12+ messages in thread
From: Daniel Jacobowitz @ 2002-01-10 11:30 UTC (permalink / raw)
To: Kevin Buettner; +Cc: Michael Elizabeth Chastain, Richard.Earnshaw, gdb
On Thu, Jan 10, 2002 at 12:09:33PM -0700, Kevin Buettner wrote:
> On Jan 10, 10:58am, Michael Elizabeth Chastain wrote:
>
> > If you build gcc and gdb in a unified build tree then dejagnu will find
> > and use the uninstalled compiler in the unified build tree, even in
> > preference to $PATH. That behaviour actually got in my way so I stopped
> > using unified trees.
>
> Here's one way to use a different compiler in a unified tree...
>
> 1) In gdb/testsuite (in your build directory), do
>
> make site.exp
>
> 2) Add lines similar to the following to the end of
> gdb/testsuite/site.exp:
>
> set CC_FOR_TARGET "/some/other/path/to/gcc"
> set CXX_FOR_TARGET "/some/other/path/to/g++"
>
> If anyone knows of other (more elegant) tricks that can be used to
> accomplish the same thing, I'd like to hear about them...
I do this by different board files.
In dejagnu/baseboards, copy unix.exp. Edit it. It should originally
say:
set_board_info compiler "[find_gcc]";
I have:
set_board_info compiler "/opt/src/gcc/install-30/bin/gcc";
set_board_info c++compiler "/opt/src/gcc/install-30/bin/g++";
There's matching variables for all the other compiler types (Java etc).
I don't yet test those.
Then, to run:
../../../src/dejagnu/runtest --target_board "unix unix3" blah.exp
That will run tests for both the default GCC and the custom in
install-30. You can also do:
--target_board "unix/gdb:debug_flags=-gdwarf-2"
to test DWARF-2, or likewise for any other flag. There's a pure
/dwarf2 option but don't use it; that's dwarf-1 level 2. I've been
forgetting to file a DejaGNU bug report about this.
The results of this thread should go in the GDB manual somewhere!
--
Daniel Jacobowitz Carnegie Mellon University
MontaVista Software Debian GNU/Linux Developer
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Changing the C/C++ compiler for gdb testsuite runs
2002-01-10 11:30 ` Daniel Jacobowitz
@ 2002-01-15 10:24 ` Daniel Jacobowitz
2002-01-15 11:18 ` Andrew Cagney
0 siblings, 1 reply; 12+ messages in thread
From: Daniel Jacobowitz @ 2002-01-15 10:24 UTC (permalink / raw)
To: gdb; +Cc: Kevin Buettner, Michael Elizabeth Chastain, Richard.Earnshaw
On Thu, Jan 10, 2002 at 02:30:40PM -0500, Daniel Jacobowitz wrote:
> I do this by different board files.
>
> In dejagnu/baseboards, copy unix.exp. Edit it. It should originally
> say:
>
> set_board_info compiler "[find_gcc]";
>
>
> I have:
> set_board_info compiler "/opt/src/gcc/install-30/bin/gcc";
> set_board_info c++compiler "/opt/src/gcc/install-30/bin/g++";
>
> There's matching variables for all the other compiler types (Java etc).
> I don't yet test those.
>
> Then, to run:
> ../../../src/dejagnu/runtest --target_board "unix unix3" blah.exp
>
> That will run tests for both the default GCC and the custom in
> install-30. You can also do:
> --target_board "unix/gdb:debug_flags=-gdwarf-2"
> to test DWARF-2, or likewise for any other flag. There's a pure
> /dwarf2 option but don't use it; that's dwarf-1 level 2. I've been
> forgetting to file a DejaGNU bug report about this.
>
> The results of this thread should go in the GDB manual somewhere!
As a followup, note that you can _NOT_ test multiple debug formats the
way that GCC tests multilibs.
You need to run with unix/gdb:debug_flags=-gdwarf-2, not just
unix/-gdwarf-2. If you do the latter, most tests will be compiled with
-g -gdwarf-2 (harmless), but nodebug.exp will be build with -gdwarf-2.
Oops.
--
Daniel Jacobowitz Carnegie Mellon University
MontaVista Software Debian GNU/Linux Developer
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Changing the C/C++ compiler for gdb testsuite runs
2002-01-15 10:24 ` Daniel Jacobowitz
@ 2002-01-15 11:18 ` Andrew Cagney
0 siblings, 0 replies; 12+ messages in thread
From: Andrew Cagney @ 2002-01-15 11:18 UTC (permalink / raw)
To: Daniel Jacobowitz
Cc: gdb, Kevin Buettner, Michael Elizabeth Chastain, Richard.Earnshaw
> The results of this thread should go in the GDB manual somewhere!
FYI, the internals section has a section on running the testsuite.
Andrew
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Changing the C/C++ compiler for gdb testsuite runs
@ 2002-01-11 8:07 Michael Elizabeth Chastain
0 siblings, 0 replies; 12+ messages in thread
From: Michael Elizabeth Chastain @ 2002-01-11 8:07 UTC (permalink / raw)
To: ac131313, Richard.Earnshaw; +Cc: gdb
I built unified trees about once per week last summer.
My experience:
+ they do solve Richard's original problem, which is:
testing a gcc without installing it
+ once I sorted out my symlink script, I had no trouble
with libiberty or include or other shared directories.
And they are synced even tighter today.
- it did bother me that I was not building the packages
in the same way that they would be released.
- I like to test with several different versions of gcc
so it's better for me to build in separate trees anyways.
Michael C
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Changing the C/C++ compiler for gdb testsuite runs
2002-01-11 7:38 ` Richard Earnshaw
@ 2002-01-11 7:45 ` Andrew Cagney
0 siblings, 0 replies; 12+ messages in thread
From: Andrew Cagney @ 2002-01-11 7:45 UTC (permalink / raw)
To: Richard.Earnshaw; +Cc: Michael Snyder, gdb
> Is the experience recent? At present mechanical processes help to
>> ensure that liberty and include are kept very much in sync (typically
>> only hours separate commits).
>>
>> Andrew
>>
>
>
> Well it can happen at any time when one product is coming off a branch and
> the other of the trunk (or a different branch). When gcc forks (or gdb
> forks) there is bound to be divergence (for obvious reasons).
Hmm, for that senario, testing GDB against an installed GCC is always best.
Andrew
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Changing the C/C++ compiler for gdb testsuite runs
2002-01-11 7:26 ` Andrew Cagney
@ 2002-01-11 7:38 ` Richard Earnshaw
2002-01-11 7:45 ` Andrew Cagney
0 siblings, 1 reply; 12+ messages in thread
From: Richard Earnshaw @ 2002-01-11 7:38 UTC (permalink / raw)
To: Andrew Cagney; +Cc: Richard.Earnshaw, Michael Snyder, gdb
>
> > But since they don't come from the same CVS tree, I can't see how to do
> > that cleanly without dicking around with the common code directories such
> > as libiberty and include -- experience has shown that with the public CVS
> > repositories that is fraught with problems. I have done it in the past,
> > but it breaks the principal that the sources you are testing are the
> > sources you are committing, so it isn't really a viable option.
>
>
> Is the experience recent? At present mechanical processes help to
> ensure that liberty and include are kept very much in sync (typically
> only hours separate commits).
>
> Andrew
>
Well it can happen at any time when one product is coming off a branch and
the other of the trunk (or a different branch). When gcc forks (or gdb
forks) there is bound to be divergence (for obvious reasons).
R.
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Changing the C/C++ compiler for gdb testsuite runs
2002-01-11 2:23 ` Richard Earnshaw
@ 2002-01-11 7:26 ` Andrew Cagney
2002-01-11 7:38 ` Richard Earnshaw
0 siblings, 1 reply; 12+ messages in thread
From: Andrew Cagney @ 2002-01-11 7:26 UTC (permalink / raw)
To: Richard.Earnshaw; +Cc: Michael Snyder, gdb
> But since they don't come from the same CVS tree, I can't see how to do
> that cleanly without dicking around with the common code directories such
> as libiberty and include -- experience has shown that with the public CVS
> repositories that is fraught with problems. I have done it in the past,
> but it breaks the principal that the sources you are testing are the
> sources you are committing, so it isn't really a viable option.
Is the experience recent? At present mechanical processes help to
ensure that liberty and include are kept very much in sync (typically
only hours separate commits).
Andrew
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Changing the C/C++ compiler for gdb testsuite runs
2002-01-10 17:30 ` Michael Snyder
@ 2002-01-11 2:23 ` Richard Earnshaw
2002-01-11 7:26 ` Andrew Cagney
0 siblings, 1 reply; 12+ messages in thread
From: Richard Earnshaw @ 2002-01-11 2:23 UTC (permalink / raw)
To: Michael Snyder; +Cc: gdb, Richard.Earnshaw
> > Is there a way of forcing the GDB testsuite to use a particular compiler
> > that hasn't yet been installed (I have a gcc-3 compiler from a build tree
> > that I'd like to use to run the tests, but I don't particularly want to
> > have to install it first). I've tried running
> >
> > make check
> >
> > with both CC_FOR_TARGET and CC set to point to the compiler, but neither
> > seems to work. There doesn't appear to be anything on this in the gdb
> > internals manual either.
>
> If you build both gcc and gdb in the same build tree,
> the tests will use the gcc from the build tree.
But since they don't come from the same CVS tree, I can't see how to do
that cleanly without dicking around with the common code directories such
as libiberty and include -- experience has shown that with the public CVS
repositories that is fraught with problems. I have done it in the past,
but it breaks the principal that the sources you are testing are the
sources you are committing, so it isn't really a viable option.
R.
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: Changing the C/C++ compiler for gdb testsuite runs
2002-01-10 5:46 Richard Earnshaw
@ 2002-01-10 17:30 ` Michael Snyder
2002-01-11 2:23 ` Richard Earnshaw
0 siblings, 1 reply; 12+ messages in thread
From: Michael Snyder @ 2002-01-10 17:30 UTC (permalink / raw)
To: gdb
Richard Earnshaw wrote:
>
> I'm not a dejagnu expert, so the answer to this may be obvious.
>
> Is there a way of forcing the GDB testsuite to use a particular compiler
> that hasn't yet been installed (I have a gcc-3 compiler from a build tree
> that I'd like to use to run the tests, but I don't particularly want to
> have to install it first). I've tried running
>
> make check
>
> with both CC_FOR_TARGET and CC set to point to the compiler, but neither
> seems to work. There doesn't appear to be anything on this in the gdb
> internals manual either.
If you build both gcc and gdb in the same build tree,
the tests will use the gcc from the build tree.
^ permalink raw reply [flat|nested] 12+ messages in thread
* Changing the C/C++ compiler for gdb testsuite runs
@ 2002-01-10 5:46 Richard Earnshaw
2002-01-10 17:30 ` Michael Snyder
0 siblings, 1 reply; 12+ messages in thread
From: Richard Earnshaw @ 2002-01-10 5:46 UTC (permalink / raw)
To: gdb; +Cc: Richard.Earnshaw
I'm not a dejagnu expert, so the answer to this may be obvious.
Is there a way of forcing the GDB testsuite to use a particular compiler
that hasn't yet been installed (I have a gcc-3 compiler from a build tree
that I'd like to use to run the tests, but I don't particularly want to
have to install it first). I've tried running
make check
with both CC_FOR_TARGET and CC set to point to the compiler, but neither
seems to work. There doesn't appear to be anything on this in the gdb
internals manual either.
R.
^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2002-01-15 19:18 UTC | newest]
Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2002-01-10 8:58 Changing the C/C++ compiler for gdb testsuite runs Michael Elizabeth Chastain
2002-01-10 11:10 ` Kevin Buettner
2002-01-10 11:30 ` Daniel Jacobowitz
2002-01-15 10:24 ` Daniel Jacobowitz
2002-01-15 11:18 ` Andrew Cagney
-- strict thread matches above, loose matches on Subject: below --
2002-01-11 8:07 Michael Elizabeth Chastain
2002-01-10 5:46 Richard Earnshaw
2002-01-10 17:30 ` Michael Snyder
2002-01-11 2:23 ` Richard Earnshaw
2002-01-11 7:26 ` Andrew Cagney
2002-01-11 7:38 ` Richard Earnshaw
2002-01-11 7:45 ` Andrew Cagney
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox