| Bug #31254 | "Max_data_length" truncated / reported wrong (compiler issue ?) | ||
|---|---|---|---|
| Submitted: | 27 Sep 2007 20:23 | Modified: | 24 Oct 2007 22:01 |
| Reporter: | Joerg Bruehe | ||
| Status: | Closed | ||
| Category: | Server: MyISAM | Severity: | S3 (Non-critical) |
| Version: | 5.1.22-rc | OS: | IBM AIX (+ i5os, 32 Bit, IBM compiler) |
| Assigned to: | Alexey Kopytov | Target Version: | |
[8 Oct 2007 15:13]
Alexey Kopytov
This is a build issue. On AIX off_t is 32-bit by default, and we have the following code
in mi_open() which truncates max_data_file_length to INT_MAX32 (0x7FFFFFFF):
#if SIZEOF_OFF_T == 4
set_if_smaller(max_data_file_length, INT_MAX32);
set_if_smaller(max_key_file_length, INT_MAX32);
#endif
We need to add -D_LARGE_FILES to CFLAGS on AIX to overcome this limitation. That makes
off_t to be defined as a signed 64-bit integer.
[12 Oct 2007 12:04]
Bugs System
A patch for this bug has been committed. After review, it may be pushed to the relevant source trees for release in the next version. You can access the patch from: http://lists.mysql.com/commits/35458 ChangeSet@1.2527, 2007-10-12 14:03:51+04:00, kaa@polly.(none) +2 -0 Fix for bug #31254: "Max_data_length" truncated / reported wrong (compiler issue ?) Problem: Improper compile-time flags on AIX prevented use of files > 2 GB. This resulted in Max_data_length being truncated to 2 GB by MyISAM code. Solution: Reverted large-file changes from the fix for bug10776. We need to define _LARGE_FILES on AIX to have support for files > 2 GB. Since _LARGE_FILE_API is incompatible with _LARGE_FILES and may be automatically defined by including standards.h, we also need a workaround to avoid this conflict.
[18 Oct 2007 23:33]
Bugs System
Pushed into 5.1.23-beta
[18 Oct 2007 23:36]
Bugs System
Pushed into 5.0.52
[24 Oct 2007 22:01]
Paul DuBois
No changelog entry needed.
Description: This report is from the 5.1.22-rc build, but the issue can also be observed in (at least) 5.1.21. It is specific to the 32 bit version on AiX and i5os, which both use the same compiler (IBM) on the PowerPC CPU. ------------------------------------------------------- *** /PATH/mysql-test/r/myisam.result --- /PATH/mysql-test/r/myisam.reject *************** *** 941,947 CREATE TABLE t1 (c1 TEXT) AVG_ROW_LENGTH=70100 MAX_ROWS=4100100100; SHOW TABLE STATUS LIKE 't1'; Name Engine Version Row_format Rows Avg_row_length Data_length Max_data_length Index_length Data_free Auto_increment Create_time Update_time Check_time Collation Checksum Create_options Comment ! t1 MyISAM 10 Dynamic X X X 72057594037927935 X X X X X X latin1_swedish_ci X max_rows=4100100100 avg_row_length=70100 DROP TABLE t1; CREATE TABLE t1 (c1 TEXT NOT NULL, KEY c1 (c1(10))) ENGINE=MyISAM; INSERT INTO t1 VALUES --- 941,947 CREATE TABLE t1 (c1 TEXT) AVG_ROW_LENGTH=70100 MAX_ROWS=4100100100; SHOW TABLE STATUS LIKE 't1'; Name Engine Version Row_format Rows Avg_row_length Data_length Max_data_length Index_length Data_free Auto_increment Create_time Update_time Check_time Collation Checksum Create_options Comment ! t1 MyISAM 10 Dynamic X X X 2147483647 X X X X X X latin1_swedish_ci X max_rows=4100100100 avg_row_length=70100 DROP TABLE t1; CREATE TABLE t1 (c1 TEXT NOT NULL, KEY c1 (c1(10))) ENGINE=MyISAM; INSERT INTO t1 VALUES ------------------------------------------------------- So the "Max_data_length" for this table is cut at 0x7FFFFFFF (max signed 32 bit) whereas the reference protocol has 0xFFFFFFFFFFFFFF (7 byte 0xFF) if I calculated it right. How to repeat: Run the test on AIX or i5os (32 bit).