Title | [INVALID] variety white-hot has asserts; hot has none |
Status | closed |
Priority | essential |
Assigned user | Richard Kistruck |
Organization | Ravenbrook |
Description | [INVALID] variety white-hot has asserts; hot has none (we suspect; not yet categorically confirmed). |
Analysis | [INVALID] this bug was never present; see analysis below. RHSK 2005-10-11 Related bugs: Note that the Variety system needs rethink, see job001158 Macros CHECK, CHECK_NONE, and CheckNONE are badly named, see job001248 This should be simple to confirm and fix. RHSK 2005-10-07 Which platform was DRJ looking at? Check comm.gmk, then check commpre/post.nmk. comm.gmk $VARIETY -> $CFxx = -DCONFIG_VAR_xx -D[nil/NDEBUG] $CFLAGS[DEBUG/OPT/OPTNODEBUG] commpre.nmk $VARIETY -> $CFxx = /DCONFIG_VAR_xx /D[nil/NDEBUG] ... config.h assume CONFIG_VAR_Hx: (line 32ff, from +CONFIG_VAR_Hx) -CONFIG_ASSERT +CHECK_DEFAULT=CheckNONE (line 49, from -CONFIG_ASSERT) +CHECK_NONE assume CONFIG_VAR_Wx; CONFIG_PROD_MPS: (line 32ff, from +CONFIG_VAR_Wx) -CONFIG_ASSERT -CHECK_DEFAULT (line 49, from -CONFIG_ASSERT) +CHECK_NONE (line 277, from +CONFIG_PROD_MPS) +PROD_CHECK_DEFAULT=CheckSHALLOW (line 301, from -CHECK_DEFAULT) +CHECK_DEFAULT=PROD_CHECK_DEFAULT=CheckSHALLOW In white-hot, +CHECK_NONE will (check.h) make all AVERs and CHECKs (but not CHECKT) be compiled away to DISCARD(). So even though white-hot MPS defines CheckSHALLOW, it should not have any assert code. In hot, +CHECK_NONE similarly makes AVERs and CHECKs compile away to nothing. It defines CHECK_DEFAULT=CheckNONE, but that has no effect. (See check.h). So, I think this job is a false alarm. I will check actual output first though. RHSK 2005-10-07. RHSK 2005-10-11 I have not checked binaries, but agreed with DRJ that his only evidence was config.h. DRJ concurs with the above analysis of config.h. So: this was never a bug. |
How found | unknown |
Evidence | DRJ looked, [at config.h -RHSK], 2005-02-25. Raw notes from Configura workshop < http://info.ravenbrook.com/mail/2005/02/28/12-40-49/0.txt >. |
Observed in | 1.105.0 |
Introduced in | 1.105.0 |
Created by | Richard Kistruck |
Created on | 2005-03-10 03:19:28 |
Last modified by | Richard Kistruck |
Last modified on | 2005-10-11 11:31:30 |
History | 2005-03-10 RHSK create 2005-10-07 RHSK start analysis 2005-10-07 RHSK looking at code suggests this is a false alarm; need to check binaries. 2005-10-11 RHSK did not check binaries, but agreed with DRJ that his only evidence was config.h. 2005-10-11 RHSK INVALID: was never a bug. Created job001248 to fix confusing macro names. |
Change | Effect | Date | User | Description |
---|---|---|---|---|
155333 | closed | 2005-10-11 11:30:23 | Richard Kistruck | MPS: empty-change to config.h, to close invalid job001157 (was never a bug) |