More record.c build failures

More record.c build failures

[Pierre Muller]

record.c is bound to give problems
on all targets that do not compile gcore.c 
source.

  gcore.o  is a optional object that
is only added to the list of compiled
object in the config directory.

  Thus it is not possible to 
use this without some configure test...

This is the outcome of
Cygwin compilation with beta cygwin 1.7:

gcc -g -O2   -Wl,--stack,8388608   \
		-o gdb.exe gdb.o libgdb.a \
		   ../readline/libreadline.a ../opcodes/libopcodes.a
../bfd/libbfd.a -lintl ../libiberty/libiberty.a
../libdecnumber/libdecnumber.a    -ltermcap -lz -lm   -liconv -lexpat
../libiberty/libiberty.a -luser32 -limagehlp gnulib/libgnu.a 
libgdb.a(record.o): In function `cmd_record_save':
/home/Famille/gdbcvs/build-norm/gdb/../../src/gdb/record.c:2223: undefined
reference to `_create_gcore_bfd'
/home/Famille/gdbcvs/build-norm/gdb/../../src/gdb/record.c:2280: undefined
reference to `_write_gcore_file'
collect2: ld returned 1 exit status
make[1]: *** [gdb.exe] Error 1

Can someone fix this, please?

Pierre Muller

Re: More record.c build failures

[Joel Brobecker]

Michael,

> Can someone fix this, please?

At this point, I think your patch should be temporary reverted until
all issues discovered thus far are solved. Would that be OK?  The meat
of the patch is still good, but it looks like little details that we
failed to notice are causing a bit of instability on the tree.

Cheers,
-- 
Joel

Re: More record.c build failures

[Tristan Gingold]

On Oct 23, 2009, at 11:46 AM, Pierre Muller wrote:

> record.c is bound to give problems
> on all targets that do not compile gcore.c
> source.
>
>  gcore.o  is a optional object that
> is only added to the list of compiled
> object in the config directory.
>
>  Thus it is not possible to
> use this without some configure test...

For the record (:-), record.c doesn't compile on Darwin because  
<byteswap.h> doesn't exists (nor
bswap_xx).

For these, maybe you could use bfd_getb64, bfd_getb32 and bfd_getb16 ?

Tristan.

Re: More record.c build failures

[Hui Zhu]

Hi Michael,

We forget some hosts don't support core dump.

I have 2 ideas to handle this issue:
1.  Make the record log don't together with core file.  Then the
record.c will not need gcore.o.
2.  Divide the "record save" cmd out of record.c.  Make gcore.o and
this .o file togther.

What do you think about them?

Thanks,
Hui

On Fri, Oct 23, 2009 at 21:37, Tristan Gingold <gingold@adacore.com> wrote:
>
> On Oct 23, 2009, at 11:46 AM, Pierre Muller wrote:
>
>> record.c is bound to give problems
>> on all targets that do not compile gcore.c
>> source.
>>
>>  gcore.o  is a optional object that
>> is only added to the list of compiled
>> object in the config directory.
>>
>>  Thus it is not possible to
>> use this without some configure test...
>
> For the record (:-), record.c doesn't compile on Darwin because <byteswap.h>
> doesn't exists (nor
> bswap_xx).
>
> For these, maybe you could use bfd_getb64, bfd_getb32 and bfd_getb16 ?
>
> Tristan.
>
>

Re: More record.c build failures

[Michael Snyder]

Pierre Muller wrote:
> record.c is bound to give problems
> on all targets that do not compile gcore.c 
> source.
> 
>   gcore.o  is a optional object that
> is only added to the list of compiled
> object in the config directory.

You're right, and I'm going to propose putting record.o
under the same sort of configury.