lib$wait from FORTRAN under VMS 8.2

[Stephen Goldberg]

We have been using lib$wait with no issues on a VMS 7.2-1 system with the same programs. We recently upgraded to VMS 8.2 and started experiencing problems. The lib$wait follows a call to a sub that uses Oracle RDB. If this routine is commented out the lib$wait works OK. If two lib$waits are used the lib$wait works OK otherwise the lib$wait acts as though the time has expired and drops through with no delay. It is like the lib$wait uses an event flag that is already in use by Oracle RDB. Any thoughts?

[Volker Halle]

Stephen,

welcome to the OpenVMS ITRC forum.

I've found an Oracle RDB V7 Release note describing this type of problem. See section

5.0.6 Application and Oracle Rdb Both Using SYS$HIBER

in the following document:

http://download-uk.oracle.com/otn_hosted_doc/rdb/pdf/rdbrn_705.pdf

Consider to use the LIB$K_NOWAKE flag to prevent spurious wakes.

Volker.

[Hein van den Heuvel]

Hmmm,

Lib$wait is documented to use SYS$HIBER
SYS$HIBER is documented to 'remember' outstanding wakes: "If one or more wakeup requests are issued for the process while it is not hibernating, the next hibernate call returns immediately; that is, the process does not hibernate. No count of outstanding wakeup requests is maintained."

I'd be tempted to toss in a extra $wake + $hiber before the lib$wait to see of this explains.

The LIB$WAIT documentation suggests to use the LIB$K_NOWAKE flag as second argument preceisely to avoid this unexepected wake problem

hth,
Hein.

[David Jones_21]

I always code expecting waits, especially sys$hiber(), as a loop with a separate predicate so they are immune to spurious wakes. I think the default behavior of LIB$WAIT was a poor choice, how many people do waits inside an AST routine?

[John Gillings]

Stephen,

This is a documented "feature" of LIB$WAIT (at least in its CRTL "sleep" incarnation). The function of LIB$WAIT can be thought of as "WAIT AT MOST" the specified time. A pending $WAKE or $WAKE during the time period will complete the LIB$WAIT prematurely.

I've attached a text file containing a substitute LIB$WAIT routine coded in MACRO32 which implements "WAIT AT LEAST" the specified time. Unlike the LIB$K_NOWAKE solution, my code will work in an AST routine (even though I would stronly recommend against any delays at AST level) or with ASTs disabled.

Note however that it will only work for an F_FLOAT argument. Modifications for other floating point types left as an exercise ;-)

[devgoswami]

I feel that the Oracle RDB subroutine which preceeds the LIB$wait call in the in line code may itself have a $ hib call inbuilt call, which causes the process to wake up in advance, before lib$wait takes effect.
If it is essential for u to use the same code structure as is , suggest using Sys$timer call, of the same time duration, as this call cannot be pre empted.
pl use the sys$bintime call to convert the required time to the language of sys$timer.
use a flag for the timer in the region of 32+ to 45, to prevent overlap with any embedded qio calls in the preceeding code section.
( as this flags are not local flags)

[Stephen Goldberg]

We have decided to use the lib$k_nowait flag. Thank you for all the help.