Page MenuHome GnuPG

compilation error libgcrypt 1.10.1
Closed, ResolvedPublic

Description

Hello,
I am trying to compile the latest release of libgcrypt for my gnupg paket.
Platform AIX 7.2, POWER, compiler IBM xlC 16.1.0.7
Using the last working compile statement, I receive the following error message:

libtool: compile:  cc -qlanglvl=extc89 -DHAVE_CONFIG_H -I. -I.. -I../src -I../src -I../mpi -I../mpi -I/opt/freeware/include -qmaxmem=16384 -DSYSV -D_AIX -D_AIX32 -D_AIX41 -D_AIX43 -D_AIX51 -D_AIX52 -D_AIX53 -D_AIX61 -D_AIX71 -D_AIX72 -D_ALL_SOURCE -DFUNCPROTO=15 -O2 -I/opt/freeware/include -c kdf.c -Wp,-qmakedep=gcc,-MF.deps/kdf.TPlo  -DPIC -o .libs/kdf.o
"kdf.c", line 955.1: 1506-343 (S) Redeclaration of _gcry_kdf_compute differs from previous declaration on line 217 of "../src/gcrypt-int.h".
"kdf.c", line 955.1: 1506-050 (I) Return type "enum {...}" in redeclaration is not compatible with the previous return type "unsigned int".

Any ideas how to resolve this?
Or can/should I stick with the LTS version of libgcrypt? (currently I am using 1.9.4)

Using the newer xlclang++ compiler throws a different failure:

libtool: compile:  xlclang++ -DHAVE_CONFIG_H -I. -I.. -I../src -I../src -I/opt/freeware/include -qmaxmem=16384 -DSYSV -D_AIX -D_AIX32 -D_AIX41 -D_AIX43 -D_AIX51 -D_AIX52 -D_AIX53 -D_AIX61 -D_AIX71 -D_AIX72 -D_ALL_SOURCE -DFUNCPROTO=15 -O2 -I/opt/freeware/include -fno-delete-null-pointer-checks -Wall -MT mpi-bit.lo -MD -MP -MF .deps/mpi-bit.Tpo -c mpi-bit.c  -DPIC -o .libs/mpi-bit.o
warning: 1540-5200 The option "-fno-delete-null-pointer-checks" is not supported.
1 warning generated.
Assembler:
.libs/mpi-bit$1.s: line 48: Error In Syntax
1500-067: (S) asm statement generates errors in assembler output.

Thank you in advance an kind regards
Frank

Details

Version
1.10.1

Event Timeline

gniibe triaged this task as Normal priority.May 17 2022, 4:31 AM
gniibe added projects: backport, Restricted Project.
gniibe added a subscriber: gniibe.

Thank you for the bug report.

For the first one, indeed, it's a bug in libgcrypt.

diff --git a/src/gcrypt-int.h b/src/gcrypt-int.h
index 08977d32..04953ffc 100644
--- a/src/gcrypt-int.h
+++ b/src/gcrypt-int.h
@@ -214,8 +214,8 @@ gpg_err_code_t _gcry_kdf_open (gcry_kdf_hd_t *hd, int algo, int subalgo,
                                const void *salt, size_t saltlen,
                                const void *key, size_t keylen,
                                const void *ad, size_t adlen);
-gcry_error_t _gcry_kdf_compute (gcry_kdf_hd_t h,
-                                const struct gcry_kdf_thread_ops *ops);
+gcry_err_code_t _gcry_kdf_compute (gcry_kdf_hd_t h,
+                                   const struct gcry_kdf_thread_ops *ops);
 gpg_err_code_t _gcry_kdf_final (gcry_kdf_hd_t h, size_t resultlen, void *result);
 void _gcry_kdf_close (gcry_kdf_hd_t h);

Fixed in master and 1.10 branch.

For the second, I wonder if newer xlclang++ compiler works with 1.9.

IIUC, it fails at libgcrypt/mpi/longlong.h, which does:

# define count_leading_zeros(count, x) \
  __asm__ ("{cntlz|cntlzw} %0,%1"                                       \
	   : "=r" ((count))                                             \
	   : "r" ((USItype)(x)))

It has two syntax-es for the instrution: one for PowerPC, another is for POWER. cntlz and cntlzw. Perhaps, this syntax is not supported by newer assembler. I suspected because in GCC, the longlong.h has been changed to only for POWER (cntlzw), ten years ago.

I do not claim I understand anything of this assembler syntax :)

For the second, I wonder if newer xlclang++ compiler works with 1.9.

I doubt it, up till now I used cc for libgcrypt and libgpg-error, because both failed with xlclang++.
The other parts gnupg, libassuan, libksba, pinentry and npth work find with xlclang++.
If you need any help in making those last two libraries work, I can try to assist.

werner changed the task status from Open to Testing.Sep 22 2022, 10:55 AM
werner removed a project: Restricted Project.