Bonjour,
I have just tried the (second) patch:
it works !
Otherwise,
do not hesitate to send me a prerelease of the upcoming version
in view to check it on diskless nodes.
Thank you very much,
Jerome
Brian W. Barrett wrote:
> On Thu, 1 May 2003, Brian W. Barrett wrote:
>
>
>>On Wed, 30 Apr 2003, Brian W. Barrett wrote:
>>
>>
>>>The info you sent earlier was of great help - i hopefully can track the
>>>problem down without too many problems. I may ask you to test a fix in
>>>the next couple days, if you don't mind.
>>
>>tkill in LAM/MPI 6.5.9 does not call fclose() on a handle it has open to
>>the killfile before it attempts to rmdir the session directory. This has
>>been fixed on the CVS head and will be fixed in the upcoming 7.0 release.
>>I've attached a 4 line patch against 6.5.9 that *should* make everything
>>work (although it is not well tested). We will probably not do a 6.5.10
>>for this bug, as we are starting to phase out the 6.5.x tree.
>>
>>Let me know if the patch does the trick for you.
>
>
> Dooh!
>
> In the "why you should always use context diffs" department, the diff I
> sent out will not work on the release 6.5.9 tarball. I have attached one
> that will. Sorry about that :).
>
>>From the top of the lam tree, 'patch -p1 < tkill-close-2.diff' should do
> the trick.
>
> Brian
>
>
>
> ------------------------------------------------------------------------
>
> *** lam-6.5.9.orig/tools/tkill/tkill.c Mon Jan 27 17:11:43 2003
> --- lam-6.5.9/tools/tkill/tkill.c Thu May 1 10:54:18 2003
> ***************
> *** 241,246 ****
> --- 241,253 ----
> if (fl_pretend) {
> exit(0);
> }
> +
> + if (fclose(fp_kill)) {
> + snprintf(extra, 1024, "fclose(for file \"%s\")", f_kill);
> + show_help(NULL, "lib-call-fail", extra, NULL);
> + exit(errno);
> + }
> +
> /*
> * Cleanup all registered objects.
> */
--
Jerome BENOIT, Ph.D.
JGMBenoit_at_[hidden]
|
