#27 closed defect (fixed)
Probe match returns wrong results
Reported by: | westram | Owned by: | westram |
---|---|---|---|
Priority: | blocker | Milestone: | |
Component: | PT server | Version: | SVN |
Keywords: | Cc: |
Description (last modified by westram)
- Results of a probe match with weighted mismatches are
- unsorted
- not filtered correctly
Occurrance:
- db: /nfshome/westram/data/5jtest.arb
- does NOT occur with gcc 3.3.3
- occurs with gcc 4.1.1 if flag -O2 or higher is used
- gcc 4.2.0 shows same behavior
Notes:
- recompiling PROBE_COM (and it's dependents) does the difference
- valgrinding ok for -O1, many warnings for -O2 (fixed)
- bond-value- is currupted:
- caused by sending doubles over AISC interface (where sth goes wrong)
- now it works again, but I'm not sure what has been the exakt cause - so the problem may appear again.
Problem was most-likely caused by type-punning (casting between very different types) w/o using a union and the use of -fstrict-aliasing (which is set by default now).
- now using -fno-strict-aliasing for AISC_COM
- several aisc-calls made typesafe by using a function call (hopefully avoiding type-punning problems)
Change History (8)
comment:1 Changed 18 years ago by westram
- Status changed from new to assigned
comment:2 Changed 18 years ago by westram
- Description modified (diff)
comment:3 Changed 18 years ago by westram
- Description modified (diff)
comment:4 Changed 18 years ago by westram
- Description modified (diff)
comment:5 Changed 18 years ago by westram
- Description modified (diff)
comment:6 Changed 18 years ago by westram
- Description modified (diff)
comment:7 Changed 18 years ago by westram
- Description modified (diff)
comment:8 Changed 18 years ago by westram
- Description modified (diff)
- Resolution set to fixed
- Status changed from assigned to closed
- not sure if really fixed, but working correctly now
Note: See
TracTickets for help on using
tickets.