ac-archive-maintainers Mailing List for Autoconf Macro Archive

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Merhaba kolay gelsin ben fort 1941 parcasilazim sizde varmı yardımcı
olursunuz teşekkür.

Hi Peter,

Peter Simons wrote:
> I feel that Steven's explanation was very insightful and
> technically accurate. I would love to see his e-mail archived, so
> that I can refer other users who have this problem to the
> message. You sent Ccs to the mailing list, so I hoped that the
> e-mail would show up there. Tom, if I may make a wish, please
> approve these e-mails.

Just did another check and they are already archived at

http://sourceforge.net/mailarchive/forum.php?forum=ac-archive-maintainers

Cheers,

--
Tom Howard

Hi Peter,

Peter Simons wrote:
> I feel that Steven's explanation was very insightful and
> technically accurate. I would love to see his e-mail archived, so
> that I can refer other users who have this problem to the
> message. You sent Ccs to the mailing list, so I hoped that the
> e-mail would show up there. Tom, if I may make a wish, please
> approve these e-mails.

I've just checked the moderation queue and there is nothing for me to 
approve.  Let's see if this one makes it into the list.

Cheers,

--
Tom Howard

On Fri, 20 Oct 2006, Steven G. Johnson wrote:

> If it fails to figure out how to link -lgfortran, then there is a bug in 
> AC_F77_LIBRARY_LDFLAGS.  However, it is more likely that you are doing the 
> wrong thing, since the code you posted contains multiple errors.

Sorry about posting erroneous code and thanks for the corrections 
(linking the same libraries twice).
I tested your corrections and it does not change anything.
I'm not going to search for a bug in fortran.m4, because I don't understand
the source-code well enough and maybe it has been fixed anyway.
(For consistency's sake I will not ship a hacked version of acx_blas.m4 
either)

> 
> > Now on openSuSE I have to do (to link with libfrtbegin and libgfortran)
> >  env LDFLAGS="-lgfortran" ./configure FC=g77
> 
> You've got it backwards.  You need to tell configure to use gfortran as 
> the F77 compiler, not g77 as the FC compiler.  i.e. do:
> 
>  	./configure F77=gfortran

Sorry about that (I thought if I'm using "AC_PROG_FC" and "F77=$FC" in 
the configure-script, I would have to do it the other way around).

 > Also, if for some reason (e.g. a bug in AC_F77_LIBRARY_LDFLAGS) you do 
> need to manually specify -lgfortran, it's not correct to put it in 
> LDFLAGS.  LDFLAGS is for linker flags (-L etc.) not for libraries.  The 
> correct thing would be to specify FLIBS, as in
> 
>  	./configure F77=gfortran FLIBS=-lgfortran

Ok, thanks. I didn't know that.

On Fri, 20 Oct 2006, Jan Wedekind wrote:
> Unfortunately I fired of my previous e-mail to early.
> I tried using Stephen's suggestion, because it makes sense to me.
> It configured successfully but when linking on openSuSE 10.0 it failed,
> because libgfortranbegin.a was not compiled with "-fPIC". On Mandriva
> on the other hand FLIBS does not contain "-lgfortran" for some reason.

If it fails to figure out how to link -lgfortran, then there is a bug in 
AC_F77_LIBRARY_LDFLAGS.  However, it is more likely that you are doing the 
wrong thing, since the code you posted contains multiple errors.

> Now on openSuSE I have to do (to link with libfrtbegin and libgfortran)
>  env LDFLAGS="-lgfortran" ./configure FC=g77

You've got it backwards.  You need to tell configure to use gfortran as 
the F77 compiler, not g77 as the FC compiler.  i.e. do:

 	./configure F77=gfortran

If you were passing FC=g77, it's no surprise that it didn't find 
-lgfortran.

Also, if for some reason (e.g. a bug in AC_F77_LIBRARY_LDFLAGS) you do 
need to manually specify -lgfortran, it's not correct to put it in 
LDFLAGS.  LDFLAGS is for linker flags (-L etc.) not for libraries.  The 
correct thing would be to specify FLIBS, as in

 	./configure F77=gfortran FLIBS=-lgfortran

Note that you should not specify environment variabls with env, you should 
specify them as arguments to the configure script.  Passing them as 
arguments to the configure script will ensure that they are correctly set 
if configure nees to be re-run via ./config.status --recheck.

>        AC_PROG_FC
>        F77=$FC
>        old_LIBS="$LIBS"
>        LIBS="$LIBS $FLIBS $FCLIBS"

This makes no sense.  First, you are adding $FLIBS and $FCLIBS to LIBS 
before running AC_F77_LIBRARY_LDFLAGS or AC_FC_LIBRARY_LDFLAGS, so those 
variables are not even set yet.  Second, ACX_LAPACK and ACX_BLAS add 
$FLIBS during linking automatically.  Third, you don't need both $FLIBS 
and $FCLIBS, since they will be identical if you set F77=$FC.

The best thing to do is to use FLIBS=@FLIBS@ in your Makefile.in and then 
use $(FLIBS) where necessary in your Makefile.  (If you use automake, then 
put $(FLIBS) in the appropriate *_LDADD variable.)

Steven

On Fri, 20 Oct 2006, Peter Simons wrote:

I didn't realize there were misunderstandings (I can also confirm that
noone as excluded in the e-mail traffic).
Here's the summary:
1) I suggested to add an attempt to link with "-lblas -lgfortran"
   at the top of the file acx_blas.m4, because this is 
2) Guido Draheim suggested to add this at the end of the file
   acx_blas.m4 and he asked me to test it.
3) Steven G. Johnson pointed out that the problem does not reside
   in acx_blas.m4.
   Instead one must choose the appropriate Fortran compiler and
   libraries (FLIBS) to successfully link with blas either by
   invoking the configure-script like this

       ./configure F77=$FC

   or with the following autoconf-code

       AC_PROG_FC
       FC=$F77
       ACX_BLAS...

Unfortunately I fired of my previous e-mail to early.
I tried using Stephen's suggestion, because it makes sense to me.
It configured successfully but when linking on openSuSE 10.0 it failed,
because libgfortranbegin.a was not compiled with "-fPIC". On Mandriva
on the other hand FLIBS does not contain "-lgfortran" for some reason.

Now on openSuSE I have to do (to link with libfrtbegin and libgfortran)
  env LDFLAGS="-lgfortran" ./configure FC=g77

And on older versions of Mandriva I have to do
  env LDFLAGS="-lgfortran" ./configure

On the newest version of Mandriva however it works.

Here's what I'm going to use in the near future for the configure-script:

  dnl For older blas-libraries you may have to use
  dnl       ./configure FC=f77
  dnl or
  dnl       ./configure FC=g77
        AC_PROG_FC
        F77=$FC
        old_LIBS="$LIBS"
        LIBS="$LIBS $FLIBS $FCLIBS"
        if test "x$F77" != "x"; then
                ACX_LAPACK([AC_DEFINE(HAVE_LIBLAPACK,yes,lapack library)]
                        [HAVE_LIBLAPACK="yes"],
                        [HAVE_LIBLAPACK="no"])
        else
                AC_MSG_WARN([Could not find Fortran compiler. Disabling 
lapack-support.])
                HAVE_LIBLAPACK="no"
        fi
        LIBS="$old_LIBS"

For the moment I'll use my proposed change to acx_blas.m4 and mark
it as a hack (at least I won't have problems with Mandriva like this):

# Try linking with gfortran and blas. Hack to provide for incomplete
# FLIBS-variable.
if test $acx_blas_ok = no; then
       save_LIBS="$LIBS"; LIBS="$LIBS -lblas -lgfortran"
       AC_MSG_CHECKING([for $sgemm in -lblas -lgfortran])
       AC_TRY_LINK_FUNC($sgemm,
       [acx_blas_ok=yes; BLAS_LIBS="-lblas -lgfortran"], [])
       AC_MSG_RESULT($acx_blas_ok)
       if test $acx_blas_ok = yes; then
          AC_MSG_WARN([Had to add -lgfortran to successfully link with 
Blas-library. The Fortran-check seems to miss out -lgfortran!])
       fi
       LIBS="$save_LIBS"
fi

Thanks for all the help. I hope it won't be a problem on recent 
distributions any more.

Regards
   Jan

(who's waiting for the day computers will be easy to use)

Hey Guido,

I apologize if I have come across as offensive. I was a little
irritated because I couldn't figure out how this e-mail thread
relates to me. I didn't know why I was receiving these e-mails. I
should have put more effort into wording my inquiries politely.

Thank you for answering my questions.

I feel that Steven's explanation was very insightful and
technically accurate. I would love to see his e-mail archived, so
that I can refer other users who have this problem to the
message. You sent Ccs to the mailing list, so I hoped that the
e-mail would show up there. Tom, if I may make a wish, please
approve these e-mails.

Have a nice day everyone,
Peter

Relax, Peter,

after getting the mail vom Jan, YOU were put always on CC,
right there from the start. There is no missing context,
just the possibility of misinterpreting the text content.
And no, I am not the list moderator, Tom is, that's why
the "stripped" mail was sent to the three of us since I
did guess that it is not in the major interest of Jan and
Steven to talk about mailing list technical details.

Is that okay with you? *g* Guido

Peter Simons wrote:
> Guido,
> 
> when you responded to my question, you have removed Steven and
> Jan from the list of recipients:
> 
>  > Peter Simons wrote:
>  >
>  >> [Articles missing in http://sourceforge.net/mailarchive/forum.php?forum_id=50529].
>  >>
>  >> Is there a technical problem?
>  >
>  > The mailinglist has just been created, probably it defaults to
>  > non-public - i.e. only subscribers may post. Perhaps the mails
>  > are stored for review by the list admin. We'll see.
> 
> I aim to put some effort into maintaining the macro archive.
> However, it is very difficult to achieve good results when I am
> thrown into conversations without context. That confuses me. I
> feel the same way about being excluded from conversations without
> knowing it; and I guess so do others. I feel it is good practice
> to retain the original list of recipients in a discussion, and to
> say so explicitly when you are adding someone or before you are
> removing someone. It helps other people to understand what is
> going on.
> 
> Regarding your response ... I am even more confused now. As far
> as I can tell, _you_ are the maintainer of that mailing list.
> When I responded, I received a message telling me that my e-mail
> awaits moderator approval. I would assume that _you_ are the
> moderator of that mailing list. Do I misunderstand something?
> 
> My perspective is that I received an e-mail in the middle of a
> discussion, in which Steven had apparently been included, because
> he was on the list of recipients. So I assumed that he was the
> author of that patch. I felt compelled to apply it and to publish
> a new version because of how the situation appeared. Jan's
> response, however, sounded more like _you_ had written the patch,
> which is why I responded again and asked for confirmation.
> Steven's response suggests that he has _not_ been included in the
> discussion, I doubt he would have recommended a patch that in his
> opinion makes no sense.
> 
> I would like to avoid this kind of misunderstanding in the
> future.
> 
> If I may make a practical suggestion, I would recommend to
> forward any feedback concerning a macro to the macro's author(s)
> and to refer the person who contacted you to them. I have found
> that all macro authors are very responsive and helpful and love
> getting feedback about their work.
> 
> With kind regards,
> Peter
> 
> 

-- 
-- guido                            http://google.de/search?q=guidod
GCS/E/S/P C++/++++$ ULHS L++w- N++@ s+:a d->++ r+@>+++ y++ (geekcode)

Guido,

when you responded to my question, you have removed Steven and
Jan from the list of recipients:

 > Peter Simons wrote:
 >
 >> [Articles missing in http://sourceforge.net/mailarchive/forum.php?forum_id=50529].
 >>
 >> Is there a technical problem?
 >
 > The mailinglist has just been created, probably it defaults to
 > non-public - i.e. only subscribers may post. Perhaps the mails
 > are stored for review by the list admin. We'll see.

I aim to put some effort into maintaining the macro archive.
However, it is very difficult to achieve good results when I am
thrown into conversations without context. That confuses me. I
feel the same way about being excluded from conversations without
knowing it; and I guess so do others. I feel it is good practice
to retain the original list of recipients in a discussion, and to
say so explicitly when you are adding someone or before you are
removing someone. It helps other people to understand what is
going on.

Regarding your response ... I am even more confused now. As far
as I can tell, _you_ are the maintainer of that mailing list.
When I responded, I received a message telling me that my e-mail
awaits moderator approval. I would assume that _you_ are the
moderator of that mailing list. Do I misunderstand something?

My perspective is that I received an e-mail in the middle of a
discussion, in which Steven had apparently been included, because
he was on the list of recipients. So I assumed that he was the
author of that patch. I felt compelled to apply it and to publish
a new version because of how the situation appeared. Jan's
response, however, sounded more like _you_ had written the patch,
which is why I responded again and asked for confirmation.
Steven's response suggests that he has _not_ been included in the
discussion, I doubt he would have recommended a patch that in his
opinion makes no sense.

I would like to avoid this kind of misunderstanding in the
future.

If I may make a practical suggestion, I would recommend to
forward any feedback concerning a macro to the macro's author(s)
and to refer the person who contacted you to them. I have found
that all macro authors are very responsive and helpful and love
getting feedback about their work.

With kind regards,
Peter

Peter Simons wrote:
> but the "ac-archive-maintainers" mailing list I found in Cc doesn't
> seem to have it:
> 
>   http://sourceforge.net/mailarchive/forum.php?forum_id=50529
> 
> The last few articles don't show up either. Is there a technical
> problem?
> 
> Anyway, it probably isn't that important.
> 

The mailinglist has just been created, probably it defaults to
non-public - i.e. only subscribers may post. Perhaps the mails
are stored for review by the list admin. We'll see.

cheerio, Guido

Thank you for the detailed response, Steven.

I feel it might be worth adding excepts from your e-mail to the
macro's documentation. This problem sounds like one others might have
too.I can't really judge it though, because I don't understand the
exact proceedings. I tried to catch up on the earlier conversation,
but the "ac-archive-maintainers" mailing list I found in Cc doesn't
seem to have it:

  http://sourceforge.net/mailarchive/forum.php?forum_id=50529

The last few articles don't show up either. Is there a technical
problem?

Anyway, it probably isn't that important.

Take care everyone,
Peter

On Wed, 18 Oct 2006, Steven G. Johnson wrote:

> Thanks for the cc.  I'm not sure the patch is the Right Thing.
> 
> The patch seems to be to check "-lblas -lgfortran", in addition to -lblas, 
> correct?  This is not the right way to get -lgfortran; it should be found 
> through FLIBS.
> 
> What's probably happening is that AC_F77_LIBRARY_LDFLAGS is finding g77 
> and using the g77 libraries, whereas you want to use gfortran.
> 
> One could, of course, use AC_FC_LIBRARY_LDFLAGS and $FCLIBS instead, which 
> should default to gfortran over g77.  (This would require changing the 
> documentation for the macro.)  However, that misses the underlying 
> problem.
> 
> The underlying problem here is that you have multiple incompatible Fortran 
> compilers installed, and if autoconf guesses the wrong one (i.e. not the 
> one that was used to compile BLAS) then linking will fail.  This can 
> happen on any system with any pair of compilers, not just g77 and 
> gfortran; e.g. if you have ifort and gfortran, or gfortran and xlf, or....
> 
> The only real solution in this kind of case is that the user has to tell 
> you which Fortran compiler to use to go with the BLAS libraries.  i.e.
> the user should use
>  	./configure F77=gfortran
> 
> To make your configure script *always* prefer the FC compiler, you can do
> 
>  	AC_PROG_FC
>  	F77=$FC
Ok, that solves the problem. Previously I was using
        env LDFLAGS="-lgfortran" ./configure
to successfully configure our software. So I mistakenly assumed, that
there was a missing test in ACX_BLAS. Sorry about trying to introduce
an inconsistency here.

Thanks a lot
Regards Jan!

Thanks for the cc.  I'm not sure the patch is the Right Thing.

The patch seems to be to check "-lblas -lgfortran", in addition to -lblas, 
correct?  This is not the right way to get -lgfortran; it should be found 
through FLIBS.

What's probably happening is that AC_F77_LIBRARY_LDFLAGS is finding g77 
and using the g77 libraries, whereas you want to use gfortran.

One could, of course, use AC_FC_LIBRARY_LDFLAGS and $FCLIBS instead, which 
should default to gfortran over g77.  (This would require changing the 
documentation for the macro.)  However, that misses the underlying 
problem.

The underlying problem here is that you have multiple incompatible Fortran 
compilers installed, and if autoconf guesses the wrong one (i.e. not the 
one that was used to compile BLAS) then linking will fail.  This can 
happen on any system with any pair of compilers, not just g77 and 
gfortran; e.g. if you have ifort and gfortran, or gfortran and xlf, or....

The only real solution in this kind of case is that the user has to tell 
you which Fortran compiler to use to go with the BLAS libraries.  i.e.
the user should use
 	./configure F77=gfortran

To make your configure script *always* prefer the FC compiler, you can do

 	AC_PROG_FC
 	F77=$FC

before calling ACX_BLAS.

Steven

The ac-archive.sf.net website has been given a new doc page,
   http://ac-archive.sourceforge.net/doc/maintainers.html
that should carry the required information pieces to do
direct maintaince work on the cvs and website repository.

Tom,

please give it a check and try to update the website
directly. Hopefully everything is in place now.

Have fun,
-- guido                            http://google.de/search?q=guidod
P.S. responses can go directly to ac-archive-maintainers list

On Wed, 18 Oct 2006, Guido Draheim wrote:

> Hi Jan,
> 
> Is it supposed to be added as the last test? After the "generic blas"
> and before the subst? I have attached a `diff -u` patch file, please
> test it on your system - it wouldn't hurt to be added to the original
> macro of S.G.J. in that position but let's be sure it works in there.

It does not really matter. The patch you send me does work for me
(openSuSE 10.0,
Thanks for your swift response.
Regards Jan!

checking for sgemm_... no
checking for ATL_xerbla in -latlas... no
checking for sgemm_ in -lblas... no
checking for sgemm_ in -lcxml... no
checking for sgemm_ in -ldxml... no
checking for sgemm_ in -lscs... no
checking for sgemm_ in -lcomplib.sgimath... no
checking for sgemm_ in -lblas... (cached) no
checking for sgemm_ in -lblas... (cached) no
checking for sgemm_ in -lblas -lgfortran... yes

-- 
Survival of the fittest is bound to brain not to money.
/---------------------------( Jan Wedekind )-------------------------\
|o.-address : Sheaf Bldg. 4314, Shef. Hallam Univ.,Pond Str., S1 1WB |
|o.-phone   : ......................... +44 (0) 114 225-3301 / -3277 |
|o.-mail/www: ................................. j.w...@sh... |
|o.-www     : ............... http://vision.eng.shu.ac.uk/mediawiki/ |
|address    : ........ Flat 7, 228-230 Langsett Rd, Sheffield S6 2UE |
|phone      : .................................. +44 (0) 114 2517186 |
|mobile     :  ................................. +44 (0) 789 6807138 |
|email/www  : .. ja...@we... / http://www.wedesoft.de[mon.co.uk] |
|SIP        : ............................ sip:72...@fw... |
|AOL        : ......................................... JanWedekind5 |
|ICQ        : ............................................ 220042632 |
|Yahoo      : ............................................... engjw3 |
|MSN        : ...................................... ja...@we... |
|gnupg      : ... 4049 5325 BE1E BC7C 6664  0256 4F88 50C5 D865 A161 |
\--------------------------------------------------------------------/

Hi Jan,

Is it supposed to be added as the last test? After the "generic blas"
and before the subst? I have attached a `diff -u` patch file, please
test it on your system - it wouldn't hurt to be added to the original
macro of S.G.J. in that position but let's be sure it works in there.

Cheerio, Guido

Jan Wedekind wrote:
> To link with blas I need libgfortran on various distributions now
> (for using blas in c++). Maybe you want to include this in
> "acx_blas.m4":
> 
> ---
> 
> 68a69,77
>> # Try linking with gfortran and blas.
>> if test $acx_blas_ok = no; then
>>       save_LIBS="$LIBS"; LIBS="$LIBS -lblas -lgfortran"
>>       AC_MSG_CHECKING([for $sgemm in -lblas -lgfortran])
>>       AC_TRY_LINK_FUNC($sgemm, [acx_blas_ok=yes; BLAS_LIBS="-lblas 
> -lgfortran"
> ], [])
>>       AC_MSG_RESULT($acx_blas_ok)
>>       LIBS="$save_LIBS"
>> fi
>>
> 
> ---
> Regards Jan!
> 

2006	Jan	Feb	Mar	Apr	May	Jun	Jul	Aug	Sep	Oct (16)	Nov	Dec
2024	Jan	Feb	Mar	Apr	May (1)	Jun	Jul	Aug	Sep	Oct	Nov	Dec

ac-archive-maintainers Mailing List for Autoconf Macro Archive

ac-archive-maintainers — Discussion list for the autoconf macro archive