[LLVMdev] [PATCH] patch to compile llvm-gcc using nightly tester script(NewNightlyTester.pl)

Tue Jul 8 11:33:41 PDT 2008

Hi Tanya,
See my comments in line,
On Tue, Jul 8, 2008 at 3:25 AM, Tanya M. Lattner <tonic at nondot.org> wrote:

>
> Rajika,
>
> A couple of comments:
>
> - You should provide a way to specify where llvm-gcc is built (just like
> llvm).

I think this is given by LLVMGCCDIR .  And also what about the location
where the actual llvm-source is kept(which will be updated by the script) ?

>
>
> - I would highly recommend allowing the user to only update llvm-gcc and
> not check it out from scratch each time. Checking out llvm-gcc is very
> time consuming.

So where should we maintain the llvm-source tree in a localmachine, so that
the script will update that copy? Are we going to get that location using an
ENV varible ? And the installtion  path will be given by LLVMGCCDIR, or to a
default location if it is not set.

You would need to make sure that llvm and llvm-gcc have
> the same rev number and nuke the llvm obj/install dirs so you get a clean
> build. Sure svn issues can happen, but they are rare and your script
> changes should be able to catch those errors.

I can keep track of the revision number of llvm and llvm-gcc once they
different from each other I can throw an error, but why should they be
synchronised to be the same revision number? I am still not clear how
that(i.e. each of the revision number in llvm and llvm-gcc of course) affect
the llvm-gcc build.

>
>
> - Remove the cvs stuff, its dead. Feel free to clean up the
> script and remove ALL cvs stuff.

Sure I'll do this and I'll be renaming all the CVS stuffs into SVN ex: the
SVN check out log is still CVS-Log.txt(it will be SVN-Log.txt etc..)

>
>
> - I think if LLVM_GCC_CONFIGURE is set and the flag -nollvmgcc is not set,
>   you should throw an error.

Well what if a user has a value for the environment varible
LLVM_GCC_CONFIGURE but he doesn't want to build the llvm-gcc tree ?
What I thought was if a user needs to build llvm-gcc he/she should set
-nollvmgcc flag, if it going to happens in the normal way (i.e. they buid
llvm, llvm-gcc, test etc.., all the time) we can simply drop this flag. I
added this following -nobuild flag for llvm.
Or simply if a user set -nobuild flag we can simply build neither llvm nor
llvm-gcc ?

>
> - You should not try to add the configure options yourself, this is not
> worth it in my opinion (maybe in the future if you want to base it off the
> target triple or something) but for now.. keep it simple. You should only
> add the --enable-llvm since only the script knows where its built.

Ok

>
>
> - What happens if LLVMGCCDIR is set?

So that , that LLVMGCCDIR will be used to install the llvm-gcc, discarding
the normal default location sepecified by the user (see above)

It will use that over the llvm-gcc
> you have built.

> In fact, I don't think you have reconfigured llvm at all
> after building llvm-gcc.

You mean to run the Deja GNU tests right? . OK I'll add that part as well

>
>
> - Checkout to something more meaningful then dst-directory.

What about llvm-gcc as Bill suggested ?

>
>
> - We don't care about warnings, file sizes or loc of llvm-gcc. If you want
>   to keep any stats you can keep track of config, build times, and build
>   status. The rest is stuff we don't really want to track in the database
>  (too much info!).

Ok, I'll not worry about LOC of llvm-gcc etc..

>
>
> Also, how have you tested this??

Yeah, I ran it (with -nosubmit) and had a look at the output.log which
indicates the values of the newly added varibles for llvm-gcc.
And I have the following TEXT files under tests results directory.

 2008-07-07-Build-Log.txt
 2008-07-07-CVS-Log.txt
 2008-07-07-Dejagnu-testrun.log
 2008-07-07-Dejagnu-testrun.sum
 2008-07-07-DejagnuTests-Log.txt
 2008-07-07-LLVMGCC-Build-Log.txt
 2008-07-07-LLVMGCC-Warnings.txt
 2008-07-07-MultiSource-Performance.txt
 2008-07-07-MultiSource-ProgramTest.txt
 2008-07-07 23:11 2008-07-07-MultiSource-Tests.txt
 2008-07-07 23:14 2008-07-07-Olden-tests.txt
 2008-07-07 23:11 2008-07-07-Performance.txt
 2008-07-07 23:04 2008-07-07-SingleSource-Performance.txt
 2008-07-07-SingleSource-ProgramTest.txt
 2008-07-07-SingleSource-Tests.txt
 2008-07-07-Tests.txt
 2008-07-07-Warnings.txt

In which LLVMGCC-* are the newly created result files for LLVM-GCC build.

Where is the patch to the DB schema and
> accept scripts?

I didn't look into the server side yet, will send that patch soon.

Thanks for the comments and help. I'll attacch the updated patch soon.

-Rajika

>
>
> -Tanya
>
> On Mon, 7 Jul 2008, Rajika Kumarasiri wrote:
>
> > hello everybody,
> >
> > I have added few improvements to my patch. Please review the new patch
> > directly.
> > Thanks!
> >
> > -Rajika
> >
> > On Sun, Jul 6, 2008 at 9:17 PM, Rajika Kumarasiri <rajikacc at gmail.com>
> > wrote:
> >
> >> hello every body.
> >>
> >> Here with I have attached the patch which compile the llvm-gcc using
> >> nightly tester script. This patch add the following capabilities to the
> >> NewNightlyTester.pl script.
> >>
> >> 1. Checkout the llvm-gcc4.2 source from the SVN.
> >> 2. Compile the checkout llvm-gcc4.2 source tree.
> >> 3. Gather the configure/make out put informations.
> >> 4. Add the (3) informations to the %hash_of_data hash to be posted to
> the
> >> server, of course for further processing.
> >>
> >> Please review the patch and let me know your comments. There are few
> small
> >> TODOs (ex: get the LOC of llvm-gcc etc..,) to be added. I'll update the
> >> patch with that soon.
> >>
> >> Thank you
> >>
> >> -Rajika
> >>
> >> --
> >> comp.lang.c - http://groups.google.com/group/comp.lang.c/topics
> >
> >
> >
> >
> >
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>

-- 
comp.lang.c - http://groups.google.com/group/comp.lang.c/topics
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20080709/f1e63dc3/attachment.html>