| Ticket UUID: | 1382528 | |||
| Title: | dict for {k v} $emptyDict crashes | |||
| Type: | Bug | Version: | obsolete: 8.5a4 | |
| Submitter: | a_kovalenko | Created on: | 2005-12-16 15:55:16 | |
| Subsystem: | 47. Bytecode Compiler | Assigned To: | dkf | |
| Priority: | 9 Immediate | Severity: | ||
| Status: | Closed | Last Modified: | 2005-12-19 05:42:40 | |
| Resolution: | Fixed | Closed By: | dkf | |
| Closed on: | 2005-12-18 22:42:40 | |||
| Description: |
Test case:
proc bad {} {
dict for {k v} {} {}
catch {whoops}
}
bad
causes segfault (linux), access violation (windows).
The crash itself appears in TclStackFree (when
decrementing refcount), but seems to be related with
Tcl_DictObjFirst / Tcl_DictObjDone behavior with empty
values.
| |||
| User Comments: |
dkf added on 2005-12-19 05:42:40:
Logged In: YES user_id=79902 I'm not going with that because it is likely to inhibit future bytecode optimizations. Instead, I've split the code so that the handling of loop-exit is different from the handling of loop-never-entered. This is a few instructions longer, but probably faster. BTW, thanks for spotting that it was a problem with the management of the catch stack; I was stumped! a_kovalenko added on 2005-12-16 23:58:55: File Added - 160100: dict-for.patch a_kovalenko added on 2005-12-16 23:58:54: Logged In: YES user_id=241496 The patch solving this problem is attached. beginCatch is now emitted before the first JumpTrue. a_kovalenko added on 2005-12-16 23:48:18: Logged In: YES user_id=241496 I know. The problem is in exception ranges: beginCatch bytecode is not reached with empty dicts, but endCatch is. dkf added on 2005-12-16 23:00:15: Logged In: YES user_id=79902 Only happens in compiled [dict for]. | |||
Attachments:
- dict-for.patch [download] added by a_kovalenko on 2005-12-16 23:58:55. [details]