| Ticket UUID: | 1116542 | |||
| Title: | bogus [binary format d] output | |||
| Type: | Bug | Version: | obsolete: 8.4.9 | |
| Submitter: | dgp | Created on: | 2005-02-04 23:03:06 | |
| Subsystem: | 12. ByteArray Object | Assigned To: | dkf | |
| Priority: | 8 | Severity: | ||
| Status: | Closed | Last Modified: | 2005-09-27 22:57:02 | |
| Resolution: | Fixed | Closed By: | dgp | |
| Closed on: | 2005-09-27 15:57:02 | |||
| Description: |
% info patch 8.4.9 % parray tcl_platform tcl_platform(byteOrder) = littleEndian tcl_platform(machine) = alpha tcl_platform(os) = Linux tcl_platform(osVersion) = 2.2.23 tcl_platform(platform) = unix tcl_platform(user) = dgp tcl_platform(wordSize) = 8 % binary scan [binary format d 83] c* v; set v 0 0 0 0 0 0 0 0 It appears the [binary format d] spits out all x00 bytes. This is fixed in 8.5a3. Can someone who knows what happened backport the fix please? | |||
| User Comments: |
dgp added on 2005-09-27 22:57:02:
Logged In: YES user_id=80530 tests fine. thanks. dkf added on 2005-09-27 22:46:05: Logged In: YES user_id=79902 Fixed in 8.4 branch I believe (and I think the HEAD was immune for unrelated reasons). Retest and reopen if the fix fails please (it works for me) dgp added on 2005-03-12 00:40:44: Logged In: YES user_id=80530 not a Linux/Alpha bug so much as brokenness of an ancient compiler. not clear what could go in compat to deal with this? dkf added on 2005-02-28 16:12:50: Logged In: YES user_id=79902 If there's a bug like this on linux-alpha, should there be updates to configure to detect it and install a compat/ workaround in that case? (Fix looks good BTW, but the test probably ought to use a value that has an exact binary representation: 1.25 instead of 1.2?) dgp added on 2005-02-26 05:22:55: File Added - 123099: 1116542.patch Logged In: YES user_id=80530 Attached patch is a partial backport from the HEAD that corrects this problem. dgp added on 2005-02-26 04:55:28: Logged In: YES user_id=80530 This is apparently a problem with memcpy() on this system. Replacing "sizeof(double)" with "1+sizeof(double)" makes things work. Leaving it as it is, only one byte is copied by memcpy() instead of the full 8 bytes of the double value. hobbs added on 2005-02-26 02:29:24: Logged In: YES user_id=72656 This works fine for me with AMD64/Linux, Sol8-64 and HP-UX 11 64-bit (pa-risc). dgp added on 2005-02-26 02:19:28: Logged In: YES user_id=80530 Bug was likely introduced with this change: 9/15/97 (bug fix) Alignment problems in "binary format" would cause a crash on some platforms when formatting floating point numbers. (SS) dgp added on 2005-02-26 02:02:34: Logged In: YES user_id=80530 Bug arrived in Tcl 8.0p1: % info patch 8.0 % binary scan [binary format d 1.2] d w; set w 1.2 % info patch 8.0p1 % binary scan [binary format d 1.2] d w; set w 2.51973479379e-322 How can we not have tests in the test suite that [binary scan] and [binary format] are inverse functions? dgp added on 2005-02-26 01:41:28: Logged In: YES user_id=80530 FWIW, this brokenness showed up sometime between Tcl 8.0 and Tcl 8.0.3. Still investigating... dkf added on 2005-02-09 21:31:28: Logged In: YES user_id=79902 What happened was TIP#129, but why that makes a difference I really don't know. | |||
Attachments:
- 1116542.patch [download] added by dgp on 2005-02-26 05:22:55. [details]