Re: ☠ Buildbot (GNU Toolchain): elfutils-try-debian-armhf - failed compile (failure) (users/marxin/try-zstd-support-v2)

["Martin Liška" <mliska@suse.cz>]

On 12/21/22 16:21, Mark Wielaard wrote:
> Hi,
> 
> On Wed, 2022-12-21 at 15:42 +0100, Mark Wielaard wrote:
>>
>> Unfortunately buildbot itself doesn't show the config.log.
>> Trying to get that...
>>
>> configure:6961: checking for library containing ZSTD_compressStream2
>> configure:6992: gcc -o conftest -D_FORTIFY_SOURCE=3 -g -O2
>> -fsanitize=undefined 
>> -fno-sanitize-recover -fsanitize=address -fno-sanitize-
>> recover   conftest.c -llz
>> ma -lbz2  >&5
>> /usr/bin/ld: /tmp/ccKZnuYA.o: in function `main':
>> /var/lib/buildbot/workers/wildebeest/elfutils-try-debian-
>> armhf/build/conftest.c:
>> 30: undefined reference to `ZSTD_compressStream2'
>> collect2: error: ld returned 1 exit status
>> configure:6992: $? = 1
>> configure: failed program was:
>>> /* confdefs.h */
>>> #define PACKAGE_NAME "elfutils"
>>> #define PACKAGE_TARNAME "elfutils"
>>> #define PACKAGE_VERSION "0.188"
>>> #define PACKAGE_STRING "elfutils 0.188"
>>> #define PACKAGE_BUGREPORT "https://sourceware.org/bugzilla"
>>> #define PACKAGE_URL "http://elfutils.org/"
>>> #define PACKAGE "elfutils"
>>> #define VERSION "0.188"
>>> #define DEFAULT_AR_DETERMINISTIC false
>>> #define HAVE_VISIBILITY 1
>>> #define HAVE_STDATOMIC_H 1
>>> #define _FILE_OFFSET_BITS 64
>>> #define CHECK_UNDEFINED 1
>>> #define USE_ZLIB 1
>>> #define USE_BZLIB 1
>>> #define USE_LZMA 1
>>> /* end confdefs.h.  */
>>>
>>> /* Override any GCC internal prototype to avoid an error.
>>>    Use char because int might match the return type of a GCC
>>>    builtin and then its argument prototype would still apply.  */
>>> #ifdef __cplusplus
>>> extern "C"
>>> #endif
>>> char ZSTD_compressStream2 ();
>>> int
>>> main ()
>>> {
>>> return ZSTD_compressStream2 ();
>>>   ;
>>>   return 0;
>>> }
>>
>> configure:6992: gcc -o conftest -D_FORTIFY_SOURCE=3 -g -O2
>> -fsanitize=undefined 
>> -fno-sanitize-recover -fsanitize=address -fno-sanitize-
>> recover   conftest.c -lzs
>> td  -llzma -lbz2  >&5
>> configure:6992: $? = 0
>> configure:7009: result: -lzstd
>>
>> ehe? So it fails and then succeeds anyway?
>>
>> Strangely the same happens for the other compression library
>> checks...
>> lzma_auto_decoder and BZ2_bzdopen
>>
>> I don't fully understand this. Oddly the same seems to happen locally
>> on my x86_64 box. So maybe the configure macros we have been using
>> have
>> always been bad/broken?
> 

Hi.

Good!

> No, this is "normal". It first tries without the -l library (which is
> the failing case) and then with the -l library (which succeeds).

Ah, makes sense.

> 
> What goes wrong is that this debian old stable arm setup has libzstd
> containing a ZSTD_compressStream2 symbol...
> 
>     40: 0000d349   264 FUNC    GLOBAL DEFAULT   11 ZSTD_compressStream2
> 
> But the zstd.h header doesn't expose it...

Yeah, that's their way of how to move an API from "staging" into a stable state.
It was changed in:
https://github.com/facebook/zstd/commit/d7d89513d6a21

and so it's present in zstd since 1.4.0. Can we somehow specify library version
in configure.ac?

Thanks,
Martin

> 
> /*******************************************************************************
> *********
>  *   ADVANCED AND EXPERIMENTAL FUNCTIONS
>  *******************************************************************************
> *********
>  * The definitions in the following section are considered experimental.
>  * They are provided for advanced scenarios.
>  * They should never be used with a dynamic library, as prototypes may change in
>  the future.
>  * Use them only in association with static linking.
>  * *****************************************************************************
> **********/
> 
> #if defined(ZSTD_STATIC_LINKING_ONLY) && !defined(ZSTD_H_ZSTD_STATIC_LINKING_ONLY)
> 
> ...
> 
> grmbl.
> 
> This is somewhat unfortunate. We need a smarter configure check for this...
> 
> Cheers,
> 
> Mark