queue-handler 40.4 problem

[Romanujan]

I think I have found a bug (regression comparing to Commodore implementation) in http://aminet.net/package/util/sys/HWGQueue - queue handler replacement, which is also a part of the unofficial OS 3.9 BoingBag 3&4. The use case is simple:

> echo >PIPE:testpipe a        ; this command executes without problems               
> echo >PIPE:testpipe a        ; doing again gives error, as you can see below
echo: unable to open redirection file

The Commodore implementation allows me to issue the second command, there is no error. Unfortunately, I can't contact the author (e-mail not valid anymore - hardly a surprise after over 20 years). Any idea? Am I trying to do something illegal?

[Minuous]

Doing a manual flush between the two invocations appears to work around the issue:

echo >PIPE:testpipe a
type PIPE:testpipe >NIL:
echo >PIPE:testpipe a

Given the reasonably small size of the executable, it might be feasible in the future to disassemble/decompile it and compare against the official version, to isolate and fix the bug, assuming all attempts to contact Heinz Wrobel fail.

[Romanujan]

This won't fix my problem - I want the first string to wait there until it is picked up by another process; not to be flushed away.

For now I had to go back to the original implementation.

[daxb]

You may have a look at AWNPipe (Aminet). At least the following can be lauched several times without errors:

echo "hello World" >AWNPipe:test

[Minuous]

BB3&4 1.37 has been released and can be downloaded from http://amigan.1emu.net/releases/#others . queue-handler 40.4 has been removed.

[Romanujan]

IMHO this is the only sane solution for now.

[HanSolo]

So which lastest version of queue handler should I copy to L: drawer ?

[Minuous]

V37.12.

[HanSolo]

I see that you updated queue-handler from 37.12 to 50.20 in current BB4 1.38. What was changed ? Who make it ?

[Romanujan]

Queue-Handler 50.20
    * Rewritten from scratch.
    * Pipe names are now limited to 107 characters.
    * All FIFO data is now allocated from memory pools to reduce
      fragmentation. It also makes cleanup a little easier.
    * The handler now defaults to blocking mode. Unless specifically
      called for, writing to a pipe that has filled up will block until
      data has been drained by reading from it.
    * Pooling FIFO data could lose bytes. Fixed.
    * Writing zero bytes to a FIFO could have strange consequences. Fixed.

So far, seems to work.

[HanSolo]

Ok but who make it ? Could this hero make more bugfixed system stuff ?

[Minuous]

I don't know who made it. I will endeavour to ascertain this information.