Bug #50356 | main.gis test fails when running 32-bit binary on Solaris on x86 | ||
---|---|---|---|
Submitted: | 15 Jan 2010 9:10 | Modified: | 12 Feb 2010 10:42 |
Reporter: | Olav Sandstå | Email Updates: | |
Status: | Duplicate | Impact on me: | |
Category: | MySQL Server: GIS | Severity: | S3 (Non-critical) |
Version: | 5.5.99-m3 (Celosia) | OS: | Solaris (x86 32-bit) |
Assigned to: | CPU Architecture: | Any | |
Tags: | Sun Studio |
[15 Jan 2010 9:10]
Olav Sandstå
[15 Jan 2010 10:29]
Sveta Smirnova
Thank you for the report. Verified as described.
[22 Jan 2010 10:59]
Olav Sandstå
The original test failure was observed when running a 32 bit binary on a Solaris 10 on an x86 server. When running the same test on a 32 bit binary on a Solaris 10 on SPARC server the test passes without any failures.
[25 Jan 2010 13:22]
Olav Sandstå
This test failure was observed having used Sun Studio 12 to compile MySQL. Using gcc 3.4.3 to compile 32-bit binaries the test runs without errors.
[25 Jan 2010 14:12]
Olav Sandstå
Summary of my experimenting with different platforms and compilers: * The test failure is only reproducible when using the Sun Studio compiler to compile 32 bit binaries and running the test on Solaris 10 on Intel or AMD x86 processors. * It is not reproducible when running on: -64 bit binaries -on SPARC -or using the GCC 3.4.3 compiler to generate 32-bit code on x86
[26 Jan 2010 11:20]
Olav Sandstå
To see if this test failure also is visible with older versions I used the latest source from mysql-5.1-bugteam and compiled it as given in the "how to repeat" section (without enabling support for DTrace). With this version of MySQL the main.gis test passes. One difference I notice between the binaries when using the file command: mysql-5.1-bugteam: mysqld: ELF 32-bit LSB executable 80386 Version 1 [FPU], dynamically linked, not stripped mysql-next-mr-bugfixing (5.5): mysqld: ELF 32-bit LSB executable 80386 Version 1 [SSE FPU], dynamically linked, not stripped I do not know the exact meaning of difference between these two but it seems to me that the 5.5 binaries are compiled to take advantage of the SSE instruction set.
[26 Jan 2010 11:37]
Olav Sandstå
To check if this test failure is present in our latest published binaries I did the following: 1. Downloaded the latest public binary of 5.5 for Solaris 32 bit on x86, unpacked it 2. Run the main.gis test. The test does not fail with this binary.
[12 Feb 2010 10:42]
Alexander Nozdrin
It's duplicate of Bug#38959.
[27 Mar 2010 19:22]
Joerg Bruehe
The same difference still happens in the release build of 5.5.3-m3, even though the fix for bub#38959 is contained. See my note about the platform difference there. The above "duplicate" info is incomplete, anyways: bug#38959 describes only the first difference. The third one was (later) reported in bug#52208. The second one was not reported anywhere else yet, but I will add it to bug#52208.