FTBFS aegis

[Walter Franzini-4]

Hi,

the last upload of aegis (4.24-5.2) fails building on hurd-i386
(sthibault) with the following errors:

------------------------------------------------------------------------
CXX="g++" /bin/sh etc/test.sh -shell /bin/sh -run \
                test/00/t0036a.sh test/00/t0036a.ES
aegis: gid 1005 unknown
FAILED test of the aenpr branch functionality (new project 117)
CXX="g++" /bin/sh etc/test.sh -shell /bin/sh -run \
                test/00/t0037a.sh test/00/t0037a.ES
aegis: warning: test mode
aegis: gid 1005 unknown
NO RESULT when testing the progress messages functionality
------------------------------------------------------------------------

It seems to me that the build is run under gid 1005 but a matching entry
is missing in /etc/group.  

Can someone check if my assumption is correct and eventually add the
missing entry into /etc/group?

Thanks
--
Walter Franzini
http://aegis.stepbuild.org/

[Samuel Thibault-8]

Walter Franzini, le Fri 16 Oct 2009 09:48:03 +0200, a écrit :
> It seems to me that the build is run under gid 1005 but a matching entry
> is missing in /etc/group.  

Ah, yes, I had propagated passwd only, not group.  I have fixed it and
requeued

Thanks!

Samuel


--
To UNSUBSCRIBE, email to debian-hurd-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Walter Franzini-4]

Samuel Thibault <sthibault@...> writes:

> Walter Franzini, le Fri 16 Oct 2009 09:48:03 +0200, a écrit :

>> It seems to me that the build is run under gid 1005 but a matching
>> entry is missing in /etc/group.
>
> Ah, yes, I had propagated passwd only, not group.  I have fixed it and
> requeued

Thanks for your prompt reply!
--
Walter Franzini
http://aegis.stepbuild.org/

[Walter Franzini-4]

Samuel Thibault <sthibault@...> writes:

> Walter Franzini, le Fri 16 Oct 2009 09:48:03 +0200, a écrit :
>> It seems to me that the build is run under gid 1005 but a matching entry
>> is missing in /etc/group.  
>
> Ah, yes, I had propagated passwd only, not group.  I have fixed it and
> requeued

The error is now more interesting:

------------------------------------------------------------------------
aegis: fcntl("/tmp/5465/lib/lockfile", F_SETLK, &{type = F_WRLCK, whence =
        SEEK_SET, start = 1, len = 1, pid = 0}): Not supported
------------------------------------------------------------------------

Any suggestion?

--
Walter Franzini
http://aegis.stepbuild.org/

[Samuel Thibault]

An updated log is available on buildd.debian.org/aegis


--
To UNSUBSCRIBE, email to debian-hurd-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Samuel Thibault-8]

Walter Franzini, le Fri 16 Oct 2009 17:31:10 +0200, a écrit :

> Samuel Thibault <sthibault@...> writes:
> > Walter Franzini, le Fri 16 Oct 2009 09:48:03 +0200, a écrit :
> >> It seems to me that the build is run under gid 1005 but a matching entry
> >> is missing in /etc/group.  
> >
> > Ah, yes, I had propagated passwd only, not group.  I have fixed it and
> > requeued
>
> The error is now more interesting:
>
> ------------------------------------------------------------------------
> aegis: fcntl("/tmp/5465/lib/lockfile", F_SETLK, &{type = F_WRLCK, whence =
> SEEK_SET, start = 1, len = 1, pid = 0}): Not supported
> ------------------------------------------------------------------------
>
> Any suggestion?

Help implementing it :)

As reported above, GNU/Hurd doesn't support record locking yet.

Samuel


--
To UNSUBSCRIBE, email to debian-hurd-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Walter Franzini-4]

Samuel Thibault <sthibault@...> writes:

[...]

>> Any suggestion?
>
> Help implementing it :)

I'm not so smart, sorry :)

> As reported above, GNU/Hurd doesn't support record locking yet.

The code contains the following fragment:

------------------------------------------------------------------------
#ifdef __hurd__
#define WHOLE_FILE_LOCKS_ONLY
#endif
------------------------------------------------------------------------

Is `__hurd__' the right macro to check?

ciao
--
Walter Franzini
http://aegis.stepbuild.org/

[Samuel Thibault-8]

Walter Franzini, le Fri 16 Oct 2009 23:31:35 +0200, a écrit :
> The code contains the following fragment:
>
> ------------------------------------------------------------------------
> #ifdef __hurd__
> #define WHOLE_FILE_LOCKS_ONLY
> #endif
> ------------------------------------------------------------------------
>
> Is `__hurd__' the right macro to check?

Nope, it's __GNU__.

But can't aegis detect that at configure time?  That will permit to
re-enable the record lock code automatically when GNU/Hurd gets the
support.

Samuel


--
To UNSUBSCRIBE, email to debian-hurd-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Samuel Thibault-8]

Walter Franzini, le Fri 16 Oct 2009 17:31:10 +0200, a écrit :
> The error is now more interesting:
>
> ------------------------------------------------------------------------
> aegis: fcntl("/tmp/5465/lib/lockfile", F_SETLK, &{type = F_WRLCK, whence =
> SEEK_SET, start = 1, len = 1, pid = 0}): Not supported
> ------------------------------------------------------------------------

The answer hasn't got to the bts: F_SETLK is not supported (yet) on
hurd-i386.  However, since the testsuite can be skipped thanks to
DEB_BUILD_OPTIONS=nocheck, I'll upload a build.

Samuel


--
To UNSUBSCRIBE, email to debian-hurd-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Walter Franzini-4]

Samuel Thibault <sthibault@...> writes:

> Walter Franzini, le Fri 16 Oct 2009 17:31:10 +0200, a écrit :
>> The error is now more interesting:
>>
>> ------------------------------------------------------------------------
>> aegis: fcntl("/tmp/5465/lib/lockfile", F_SETLK, &{type = F_WRLCK, whence =
>> SEEK_SET, start = 1, len = 1, pid = 0}): Not supported
>> ------------------------------------------------------------------------
>
> The answer hasn't got to the bts: F_SETLK is not supported (yet) on
> hurd-i386.  However, since the testsuite can be skipped thanks to
> DEB_BUILD_OPTIONS=nocheck, I'll upload a build.

:-(

In 4.24.2-1 (it took me some time to find a sponsor) I've fixed the way
hurd is recognized and the tests result passed from:

------------------------------------------------------------------------
Passed 49 of 277 tests.
Failed 10 of 277 tests.
No result for 218 of 277 tests.
------------------------------------------------------------------------

to

------------------------------------------------------------------------
Passed 282 of 283 tests.
Failed 1 of 283 tests.
------------------------------------------------------------------------

The remaining problem is similar, see below, so I expect to identify
where the code needs to be fixed soon (for a reasonable definition of
soon).

------------------------------------------------------------------------
aelock: fcntl("/tmp/28008/lib/lockfile", F_GETLK, &{type = F_WRLCK, whence =
        SEEK_SET, start = 0, len = 458752, pid = 0}): Function not implemented
aelock: command "aelock" exit status 1
FAILED test of the aelock functionality (develop begin 152)
------------------------------------------------------------------------

Aegis test-suite should really be run since failures means something
will not work as the *user* expects.  

While upstream develops running the tests very often, it is possible
that uncommon architectures or operating systems exposes problems.  I'm
part of the upstream team and one of the resons for taking over aegis
packages is to test the stable branch on the many architectures Debian
supports, so please do not disable the test-suite.

Ciao :-)
--
Walter Franzini
http://aegis.stepbuild.org/

[Michael Banck]

On Wed, Nov 11, 2009 at 09:10:19AM +0100, Walter Franzini wrote:

> The remaining problem is similar, see below, so I expect to identify
> where the code needs to be fixed soon (for a reasonable definition of
> soon).
>
> ------------------------------------------------------------------------
> aelock: fcntl("/tmp/28008/lib/lockfile", F_GETLK, &{type = F_WRLCK, whence =
> SEEK_SET, start = 0, len = 458752, pid = 0}): Function not implemented
> aelock: command "aelock" exit status 1
> FAILED test of the aelock functionality (develop begin 152)
> ------------------------------------------------------------------------

I believe this is real record locking (len >= 1), so something which is
not implemented yet for the Hurd.  If len 458752 is the whole file, it
could be emulated as a file-wide lock, but as it is called above, glibc
will return function not implemented.


cheers,

Michael


--
To UNSUBSCRIBE, email to debian-hurd-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...

[Walter Franzini-4]

reopen 551094
stop

Michael Banck <mbanck@...> writes:

[...]

>> ------------------------------------------------------------------------
>> aelock: fcntl("/tmp/28008/lib/lockfile", F_GETLK, &{type = F_WRLCK, whence =
>> SEEK_SET, start = 0, len = 458752, pid = 0}): Function not implemented
>> aelock: command "aelock" exit status 1
>> FAILED test of the aelock functionality (develop begin 152)
>> ------------------------------------------------------------------------
>
> I believe this is real record locking (len >= 1), so something which is
> not implemented yet for the Hurd.  If len 458752 is the whole file, it
> could be emulated as a file-wide lock, but as it is called above, glibc
> will return function not implemented.

I've finally updated the package and it still fails:

------------------------------------------------------------------------
CXX="g++" /bin/sh etc/test.sh -shell /bin/sh -run \
                test/02/t0218a-pmil.sh test/02/t0218a-pmil.ES
aelock: aelock -list
aelock: fcntl("/tmp/22868/lib/lockfile", F_GETLK, &{type = F_WRLCK, whence =
        SEEK_SET, start = 0, len = 0, pid = 0}): Function not implemented
aelock: command "aelock" exit status 1
FAILED test of the aelock functionality (develop begin 152)
------------------------------------------------------------------------

Now, if I'm not wrong, the lock is for the whole file, but it's still
unimplemented.

I'm still interested in fixing this problem, but needs some help.
Anyone can help?

thanks
--
Walter Franzini
http://aegis.stepbuild.org/

[Samuel Thibault-8]

Walter Franzini, le Sat 19 Dec 2009 22:29:39 +0100, a écrit :

> ------------------------------------------------------------------------
> CXX="g++" /bin/sh etc/test.sh -shell /bin/sh -run \
> test/02/t0218a-pmil.sh test/02/t0218a-pmil.ES
> aelock: aelock -list
> aelock: fcntl("/tmp/22868/lib/lockfile", F_GETLK, &{type = F_WRLCK, whence =
> SEEK_SET, start = 0, len = 0, pid = 0}): Function not implemented
> aelock: command "aelock" exit status 1
> FAILED test of the aelock functionality (develop begin 152)
> ------------------------------------------------------------------------
>
> Now, if I'm not wrong, the lock is for the whole file, but it's still
> unimplemented.

The F_GETLK operation is not yet implemented indeed.  Not much to do but
implement it.

Samuel


--
To UNSUBSCRIBE, email to debian-hurd-REQUEST@...
with a subject of "unsubscribe". Trouble? Contact listmaster@...