Error log not flushed before dbReadTest exits

[Richard Fearn]

dbReadTest gives rather unhelpful output for one of my db files:

$ dbReadTest db/dbExample1.db
For input file db/dbExample1.db$

Adding:

  errlogFlush();

at the end of the main function in dbReadTest.c results in more output:

$ dbReadTest db/dbExample1.db
For input file db/dbExample1.dbBad character '$' in record name "$(user):aiExample"
Record "$(user):aiExample" is of unknown type "ai"
Error at or before ")" in file "db/dbExample1.db" line 1
filename="../dbReadTest.c" line number=77
from dbReadDatabase
$

Not sure if this is the right solution, but it helps...

[mdavidsaver (mdavidsaver)]

I'm marking this bug as 3.14 series as it seems that dbReadTest is removed before 3.15.0.1. This removal my be a seperate issue.

[mdavidsaver (mdavidsaver)]

Instead for errlogFlush() can you try epicsExit()? The exit handler for errlog should flush the message buffer.

[mdavidsaver (mdavidsaver)]

I can't make dbReadTest lose parts of an error message, but I see no reason why it couldn't. Adding epicsExit(0) after dbFreeBase() does noticeable increases the runtime.

[Richard Fearn (richardfearn)]

> I'm marking this bug as 3.14 series as it seems that dbReadTest is removed before 3.15.0.1. This removal my be a seperate issue.

Yes, it was removed here: http://bazaar.launchpad.net/~epics-core/epics-base/3.15/revision/12271

[Richard Fearn (richardfearn)]

> Instead for errlogFlush() can you try epicsExit()? The exit handler for errlog should flush the message buffer.

Yeah, that also seems to guarantee that the errors are displayed:

$ dbReadTest db/dbExample1.db
For input file db/dbExample1.dbfilename="../dbLexRoutines.c" line number=240
No such file or directory dbRead opening file db/dbExample1.db
filename="../dbReadTest.c" line number=77
from dbReadDatabase

[Richard Fearn (richardfearn)]

> I can't make dbReadTest lose parts of an error message, but I see no reason why it couldn't.

Do you agree there is a problem? I noticed you marked this bug as invalid...

Then again the whole issue may not matter much if the program has been removed in 3.15.

> Adding epicsExit(0) after dbFreeBase() does noticeable increases the runtime.

It does. But slow and useful is better than fast and unhelpful :-)

[Ralph Lange (ralph-lange)]

The bug is correctly marked 'invalid' for the main trunk of development, as the test program showing the bug has been removed.
It is still there, marked 'new', for the 3.14 branch.

Orderly shutdown of the IOC takes noticeably longer.
You can also experience that effect comparing typing 'exit' on an iocShell with a simple ^C.

[Richard Fearn (richardfearn)]

> The bug is correctly marked 'invalid' for the main trunk of development, as the test program showing the bug has been removed.
It is still there, marked 'new', for the 3.14 branch.

Ah, thanks! I didn't realise it could have a different status for each EPICS version. (Haven't used Launchpad much before.)

[mdavidsaver (mdavidsaver)]

After some thought, the correctness of epicsExit() isn't worth the reduction in speed. So I just added errlogFlush() as you suggest. Also pulled in a patch dbPvdLib.c to fix this error, which I see when running dbReadTest.

> epicsMutex pthread_mutex_destroy failed: error Device or resource busy