MySQL Bugs: #28381: Building the "embedded" server fails, due to a recent code change

Bug #28381	Building the "embedded" server fails, due to a recent code change
Submitted:	11 May 2007 15:37	Modified:	8 Aug 2007 16:28
Reporter:	Joerg Bruehe	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Server: Compiling	Severity:	S1 (Critical)
Version:	5.1.18	OS:	Linux (ICC / IA64)
Assigned to:	Antony Curtis	CPU Architecture:	Any

Description:
According to the build log, the problem is caused by the changes for WL#2936.

Comparing the log of the last working (test) to the first failing (release) build, there are many new warnings of this type:

> ha_innodb.cc(178): warning #411: class "<unnamed>" defines no constructor to initialize the following:
>             const member "<unnamed>::def_val"
>   static MYSQL_THDVAR_BOOL(support_xa, PLUGIN_VAR_OPCMDARG,
>          ^
>

Much later, linking "mysqltest_embedded" fails with this message:

> /usr/lib/libc_nonshared.a(elf-init.oS)(.text+0x31): In function `__libc_csu_init':
> : relocation truncated to fit: GPREL22 __init_array_start

This breaks both the tar.gz and the RPM build.
A test build skipping the embedded server succeeded in linking,
but I did not run tests.

How to repeat:
Build unsing the Intel compiler for the IA64 CPU.

I see the compiler warnings - in fact there are a lot of compiler warnings - in particular warnings against gcc-specific syntax such as the 'pragma implementation' etc.but every attempt to build completes successfully.

However. embedded test run failes - looks like system vars not initialised correctly as results are consistantly missing system vars whenever 'SHOW VARIABLES' is executed.

Is this the faulure that needs to be fixed?

In the release builds, I don't get so far as to try the tests - linking fails.

Look at the file with the log differences
(left: working test build / right: failing release build, after the push):
    rx2620b-diff-work-fail.log

The warnings about the missing constructor are the new ones (from the push),
and I assume it it this which causes the final link failure.
I agree that the linker error message is not very helpful.

When you write "building completes", are you sure you configured using "--with-embedded-server" ?
In the release builds, this option is what caused it to fail.
Using "--without-embedded-server", I also could build - but that is not the configuration to be published.

I have been doing some research and it appears that this error message is an old bug in GNU binutils which was much discussed during 2003.

On the machine in question, I can see...

[mysqldev@rx2620b mysqldev]$ ld --version
GNU ld version 2.14.90.0.4 20030523

It is quite possible that gnu binutils needs to be upgraded on this machine.

I checked the RedHat web site:
they have binutils-2.14.90.0.4-42 for this platform,
our machine is running binutils-2.14.90.0.4-39.

I filed a sys request about an update.

The machine has bee upgraded to 
   binutils-2.14.90.0.4-42
but the linker error persists.

That binutils version is the highest one available for RedHat EL AS 3,
so there is no chance for a further upgrade.

Ok, I have examined this bug.... and it is definitely the bug reported
at http://sources.redhat.com/ml/binutils/2003-11/msg00054.html

However, I was successfully able to compile the code...

> [mysqldev@rx2620b examples]$ PATH=/users/acurtis/ia64tools/bin:$PATH
> [mysqldev@rx2620b examples]$ export PATH
> [mysqldev@rx2620b examples]$ make
> make  all-am
> make[1]: Entering directory `/data/home/mysqldev/tmp-200705311919-5.1.20-beta-17729/rx2620b-icc-glibc23/mysql-5.1.20-beta/libmysqld/examples'
> /bin/sh ../../libtool --preserve-dup-deps --tag=CXX --mode=link icpc -static-libgcc -static-libcxa -i-static  -O3 -unroll2 -ip -mp -restrict -no-ftz   -fno-implicit-templates -fno-exceptions -fno-rtti  -rdynamic  -o mysql_client_test_embedded  mysql_client_test.o ../libmysqld.a -ldl    -lpthread -lrt -lcrypt -lnsl -lm  -lpthread   -lrt -lcrypt -lnsl -lm   ../../zlib/libzlt.la
> icpc -static-libgcc -static-libcxa -i-static -O3 -unroll2 -ip -mp -restrict -no-ftz -fno-implicit-templates -fno-exceptions -fno-rtti -rdynamic -o mysql_client_test_embedded mysql_client_test.o  ../libmysqld.a -ldl -lpthread -lrt -lcrypt -lnsl -lm -lpthread -lrt -lcrypt -lnsl -lm ../../zlib/.libs/libzlt.a -lrt -lcrypt -lnsl -lm
> if icpc -static-libgcc -static-libcxa -i-static -DEMBEDDED_LIBRARY -I. -I. -I../.. -I../../include -I../../include -I. -I../.. -I../../client -I../../regex     -O3 -unroll2 -ip -mp -restrict -no-ftz   -fno-implicit-templates -fno-exceptions -fno-rtti -MT mysql.o -MD -MP -MF ".deps/mysql.Tpo" -c -o mysql.o mysql.cc; \
> then mv -f ".deps/mysql.Tpo" ".deps/mysql.Po"; else rm -f ".deps/mysql.Tpo"; exit 1; fi
> if icpc -static-libgcc -static-libcxa -i-static -DEMBEDDED_LIBRARY -I. -I. -I../.. -I../../include -I../../include -I. -I../.. -I../../client -I../../regex     -O3 -unroll2 -ip -mp -restrict -no-ftz   -fno-implicit-templates -fno-exceptions -fno-rtti -MT readline.o -MD -MP -MF ".deps/readline.Tpo" -c -o readline.o readline.cc; \
> then mv -f ".deps/readline.Tpo" ".deps/readline.Po"; else rm -f ".deps/readline.Tpo"; exit 1; fi
> if icpc -static-libgcc -static-libcxa -i-static -DEMBEDDED_LIBRARY -I. -I. -I../.. -I../../include -I../../include -I. -I../.. -I../../client -I../../regex     -O3 -unroll2 -ip -mp -restrict -no-ftz   -fno-implicit-templates -fno-exceptions -fno-rtti -MT completion_hash.o -MD -MP -MF ".deps/completion_hash.Tpo" -c -o completion_hash.o completion_hash.cc; \
> then mv -f ".deps/completion_hash.Tpo" ".deps/completion_hash.Po"; else rm -f ".deps/completion_hash.Tpo"; exit 1; fi
> /bin/sh ../../libtool --preserve-dup-deps --tag=CXX --mode=link icpc -static-libgcc -static-libcxa -i-static  -O3 -unroll2 -ip -mp -restrict -no-ftz   -fno-implicit-templates -fno-exceptions -fno-rtti  -rdynamic  -o mysql  mysql.o readline.o completion_hash.o ../../cmd-line-utils/readline/libreadline.a -lncurses ../libmysqld.a -ldl    -lpthread -lrt -lcrypt -lnsl -lm  -lpthread   -lrt -lcrypt -lnsl -lm   ../../zlib/libzlt.la
> icpc -static-libgcc -static-libcxa -i-static -O3 -unroll2 -ip -mp -restrict -no-ftz -fno-implicit-templates -fno-exceptions -fno-rtti -rdynamic -o mysql mysql.o readline.o completion_hash.o  ../../cmd-line-utils/readline/libreadline.a -lncurses ../libmysqld.a -ldl -lpthread -lrt -lcrypt -lnsl -lm -lpthread -lrt -lcrypt -lnsl -lm ../../zlib/.libs/libzlt.a -lrt -lcrypt -lnsl -lm
> make[1]: Leaving directory `/data/home/mysqldev/tmp-200705311919-5.1.20-beta-17729/rx2620b-icc-glibc23/mysql-5.1.20-beta/libmysqld/examples'
> [mysqldev@rx2620b examples]$ ./mysql_client_test_embedded
> mysql_embedded: Unknown error 1146
> 070608 12:39:45  InnoDB: Operating system error number 2 in a file operation.
> InnoDB: The error means the system cannot find the path specified.
> InnoDB: If you are installing InnoDB, remember that you must create
> InnoDB: directories yourself, InnoDB does not create them.
> InnoDB: File name /usr/local/mysql/data/ibdata1
> InnoDB: File operation call: 'create'.
> InnoDB: Cannot continue operation.
> [mysqldev@rx2620b examples]$ ld --version
> GNU ld version 2.17
> Copyright 2005 Free Software Foundation, Inc.
> This program is free software; you may redistribute it under the terms of
> the GNU General Public License.  This program has absolutely no warranty.

GNU ld version 2.17 appears to be able to link it just fine.

As best as I can tell (looking at changelogs), the bug is a linker error which was fixed by GNU ld version 2.15... so I cannot reasonably suggest any solution which does not involve moving away from ld ver 2.14.x

No feedback was provided for this bug for over a month, so it is
being suspended automatically. If you are able to provide the
information that was originally requested, please do so and change
the status of the bug back to "Open".

The build host as a whole could not be upgraded,
but a newer version of "ld" (2.17 ?) is installed below "/usr/local",
and local configuration files have been changed to use it when a server is built.

This way, linking succeeds.