[gmp] WIP

[dg0yt]

[Neumann-A]

Hmm ok that is exactly the same error I observed when trying to update gmp. Seems like something is missing here:

./libtool: eval: line 10827: `dumpbin.exe -symbols -headers .libs/libgmp.la-2.obj  | <something is missing here> | /usr/bin/sed -e ... 

[Neumann-A]

https://lists.gnu.org/archive/html/bug-libtool/2015-07/msg00000.html

[dg0yt]

@Neumann-A https://github.com/microsoft/vcpkg/issues/36639

[dg0yt]

https://lists.gnu.org/archive/html/bug-libtool/2015-07/msg00000.html

This old message probably doesn't help. config.log shows that there is still the output which should enable "MS dumpbin" format, but it doesn't for gmp. (It does for another failed package in CI, that's where I got the expected strings from.)

[Neumann-A]

Found an old build log from the vcpkg-make PR. It shos the missing piece

libtool: link: dumpbin.exe -symbols -headers @.libs\\libgmp.la.nm  | gawk '     {last_section=section; section=$ 3};     /^COFF SYMBOL TABLE/{for(i in hide) delete hide[i]};     /Section length .*#relocs.*(pick any)/{hide[last_section]=1};     /^ *Symbol name *: /{split($ 0,sn,":"); si=substr(sn[2],2)};     /^ *Type *: code/{print "T",si,substr(si,length(prfx))};     /^ *Type *: data/{print "I",si,substr(si,length(prfx))};     $ 0!~/External *\|/{next};     / 0+ UNDEF /{next}; / UNDEF \([^|]\)*()/{next};     {if(hide[section]) next};     {f="D"}; $ 0~/\(\).*\|/{f="T"};     {split($ 0,a,/\||\r/); split(a[2],s)};     s[1]~/^[@?]/{print f,s[1],s[1]; next};     s[1]~prfx {split(s[1],t,"@"); print f,t[1],substr(t[1],length(prfx))}     ' prfx=^ | /usr/bin/sed '/ __gnu_lto/d' | /usr/bin/sed -e '/^[BCDGRS][ ]/s/.*[ ]\([^ ]*\)/\1,DATA/' | /usr/bin/sed -e '/^[AITW][ ]/s/.*[ ]//' | sort | uniq > .libs/gmp.exp
libtool: link: if test DEF = "`/usr/bin/sed -n -e 's/^[	 ]*//' -e '/^\(;.*\)*$/d' -e 's/^\(EXPORTS\|LIBRARY\)\([	 ].*\)*$/DEF/p' -e q .libs/gmp.exp`" ; then cp ".libs/gmp.exp" ".libs/gmp-10.dll.def"; echo ".libs\\gmp-10.dll.def" > ".libs/gmp-10.dll.exp"; else /usr/bin/sed -e 's/^/-link -EXPORT:/' < .libs/gmp.exp > .libs/gmp-10.dll.exp; fi

[dg0yt]

It is in configure ... Can you manually reproduce those steps in configure? I don't have MSVC...

[dg0yt]

I mean: please look for lt_cv_sys_global_symbol_pipe.

[Neumann-A]

generated libtool diff: work vs broke libtool.txt

It somehow identifies it as BSD nm ? I'll try to check MSYS_ROOT next

[Neumann-A]

work vs broke. So configure is not using dumpbin ?

@@ -19109,9 +18738,11 @@ gmp_cv_option_alloca=alloca
 gmp_cv_prog_exeext_for_build=.exe
 gmp_cv_prog_m4=m4
 lt_cv_ar_at_file=@
-lt_cv_deplibs_check_method=pass_all
+lt_cv_deplibs_check_method='file_magic ^x86 archive import|^x86 DLL'
+lt_cv_file_magic_cmd=func_win32_libid
+lt_cv_file_magic_test_file=
 lt_cv_ld_reload_flag=-r
-lt_cv_nm_interface='MS dumpbin'
+lt_cv_nm_interface='BSD nm'
 lt_cv_objdir=.libs
 lt_cv_path_LD='link.exe -verbose'
 lt_cv_path_LDCXX=
@@ -19128,11 +18759,11 @@ lt_cv_prog_compiler_static_works_CXX=yes
 lt_cv_prog_gnu_ld=no
 lt_cv_prog_gnu_ldcxx=
 lt_cv_sharedlib_from_linklib_cmd=func_cygming_dll_for_implib_fallback
-lt_cv_sys_global_symbol_pipe='gawk '\''     {last_section=section; section=$ 3};     /^COFF SYMBOL TABLE/{for(i in hide) delete hide[i]};     /Section length .*#relocs.*(pick any)/{hide[last_section]=1};     /^ *Symbol name *: /{split($ 0,sn,":"); si=substr(sn[2],2)};     /^ *Type *: code/{print "T",si,substr(si,length(prfx))};     /^ *Type *: data/{print "I",si,substr(si,length(prfx))};     $ 0!~/External *\|/{next};     / 0+ UNDEF /{next}; / UNDEF \([^|]\)*()/{next};     {if(hide[section]) next};     {f="D"}; $ 0~/\(\).*\|/{f="T"};     {split($ 0,a,/\||\r/); split(a[2],s)};     s[1]~/^[@?]/{print f,s[1],s[1]; next};     s[1]~prfx {split(s[1],t,"@"); print f,t[1],substr(t[1],length(prfx))}     '\'' prfx=^ | /usr/bin/sed '\''/ __gnu_lto/d'\'''
-lt_cv_sys_global_symbol_to_c_name_address='/usr/bin/sed -n -e '\''s/^I .* \(.*\)$/  {"\1", (void *) 0},/p'\'' -e '\''s/^: \(.*\) .*$/  {"\1", (void *) 0},/p'\'' -e '\''s/^[ABCDGISTW][ABCDGISTW]* .* \(.*\)$/  {"\1", (void *) \&\1},/p'\'''
-lt_cv_sys_global_symbol_to_c_name_address_lib_prefix='/usr/bin/sed -n  -e '\''s/^I .* \(lib.*\)$/  {"\1", (void *) 0},/p'\''  -e '\''s/^I .* \(.*\)$/  {"lib\1", (void *) 0},/p'\'' -e '\''s/^: \(.*\) .*$/  {"\1", (void *) 0},/p'\'' -e '\''s/^[ABCDGISTW][ABCDGISTW]* .* \(lib.*\)$/  {"\1", (void *) \&\1},/p'\'' -e '\''s/^[ABCDGISTW][ABCDGISTW]* .* \(.*\)$/  {"lib\1", (void *) \&\1},/p'\'''
-lt_cv_sys_global_symbol_to_cdecl='/usr/bin/sed -n -e '\''s/^I .* \(.*\)$/extern __declspec(dllimport) char \1;/p'\'' -e '\''s/^T .* \(.*\)$/extern int \1();/p'\'' -e '\''s/^[ABCDGISTW][ABCDGISTW]* .* \(.*\)$/extern char \1;/p'\'''
-lt_cv_sys_global_symbol_to_import='/usr/bin/sed -n -e '\''s/^I .* \(.*\)$/\1/p'\'''
+lt_cv_sys_global_symbol_pipe=
+lt_cv_sys_global_symbol_to_c_name_address='/usr/bin/sed -n -e '\''s/^: \(.*\) .*$/  {"\1", (void *) 0},/p'\'' -e '\''s/^[ABCDGISTW][ABCDGISTW]* .* \(.*\)$/  {"\1", (void *) \&\1},/p'\'''
+lt_cv_sys_global_symbol_to_c_name_address_lib_prefix='/usr/bin/sed -n -e '\''s/^: \(.*\) .*$/  {"\1", (void *) 0},/p'\'' -e '\''s/^[ABCDGISTW][ABCDGISTW]* .* \(lib.*\)$/  {"\1", (void *) \&\1},/p'\'' -e '\''s/^[ABCDGISTW][ABCDGISTW]* .* \(.*\)$/  {"lib\1", (void *) \&\1},/p'\'''
+lt_cv_sys_global_symbol_to_cdecl=
+lt_cv_sys_global_symbol_to_import=

[Neumann-A]

Configure: Broke:

else case e in #(
  e) lt_cv_nm_interface="BSD nm"
  echo "int some_variable = 0;" > conftest.$ac_ext

Working:

else $as_nop
  lt_cv_nm_interface="BSD nm"
  echo "int some_variable = 0;" > conftest.$ac_ext

looks like a typo in how configure gets generated.

[dg0yt]

Maybe some change in autoconf2.72 vs. autoconf2.71?

[Neumann-A]

Here is the diff autoconf-diff.txt There are changes like:

-[m4_append_uniq([_AS_CLEANUP], [AS_REQUIRE([_AS_EMPTY_ELSE_PREPARE])])]dnl
-[else $as_nop
-  $1
+[else case e in @%:@(
+  e) $1 ;;
+esac

in there but I couldn't pin that to an upstream commit yet

[dg0yt]

BTW upstream discourages running autoreconf, https://gmplib.org/devel/repo-usage

[Neumann-A]

BTW upstream discourages running autoreconf Do not use autoreconf; it will overwrite config.guess which in turn will cause any builds to be awful.

They are just claiming that their config.guess is better than upstreams config.guess. That might be or might not be the case. But the problem here is unrelated to gmp. Shouldn't all make based builds which build dynamic libs currently fail due to the wrong detection?

[Neumann-A]

So I assume the MSYS update also broke xcb-util-errors but I already have a fix for that in the meson pr

[dg0yt]

And should CI have failed before the merge? Again, the config.log shows that the checks for MS dumpbin format are run, it shows the MS dumpbin output, the output shows a string

009 00000000 SECT3  notype       External     | some_variable

which should match the expected pattern in

  cat conftest.out >&5
  if $GREP 'External.*some_variable' conftest.out > /dev/null; then
    lt_cv_nm_interface="MS dumpbin"
  fi

so this where something goes wrong when we leave with

configure:12427: result: BSD nm

[Neumann-A]

A wait. You know what the problem in xcb-util-errors was? It is $(AM_V_GEN)$(PYTHON) $(srcdir)/src/extensions.py [email protected] ${XCBPROTO_XCBINCLUDEDIR}/*.xml && \ where * got directly expanded into one of the filenames in that folder instead of correctly being passed on. Maybe the same is happening here that in if $GREP 'External.*some_variable' conftest.out > /dev/null the * gets expanded to nothing instead of being passed on as a regex character.

[Neumann-A]

So strange observation: The tree which failed previously worked miraculously and was broken again by re-running the bootstrap script. Couldn't get it to work again. cc @BillyONeal

Maybe relevant: https://stackoverflow.com/questions/11456403/stop-shell-wildcard-character-expansion

[Neumann-A]

ok seems like I did a git pull in-between which of course breaks it. (I have too many vcpkg folders o.O)