Hi Phil,
On Fri, Nov 7, 2014 at 3:42 AM, p...@highoctane.be <p...@highoctane.be>
wrote:
> Code (Transcript calls of mine)
>
>
> open: fileName forWrite: writeMode
> "Open the file with the given name. If writeMode is true, allow writing,
> otherwise open the file in read-only mode."
> | f |
> f := fileName asVmPathName.
> Transcript log: 'File:', f.
>
> fileID := StandardFileStream retryWithGC:[self primOpen: f writable:
> writeMode]
> until:[:id| id notNil]
> forFileNamed: fileName.
> fileID ifNil: [
> ' > Cannot get fileID' logCr.
> ^ nil]. "allows sender to detect failure"
> (' > Got fileID: ', fileID printString) logCr.
> name := fileName.
> self register.
> rwmode := writeMode.
> buffer1 := String new: 1.
> self enableReadBuffering
>
> I have a ton of files to open. And some were slow to return a problem.
>
> After digging, I got to the method above.
>
> Now at one moment, I get nil fileIDs.
>
> What is that retryWithGC: until: thing???
>
> retryWithGC: execBlock until: testBlock forFileNamed: fullName
> "Re-implemented to only force GC if a file with the given name exists" |
> blockValue foundIt | blockValue := execBlock value. (testBlock value:
> blockValue) ifTrue:[^blockValue]. "See if we have a file with the given
> name" foundIt := self registry keys "hold on strongly for now"
> anySatisfy:[:file| file name sameAs: fullName]. foundIt
> ifFalse:[^blockValue]. Smalltalk garbageCollectMost. blockValue :=
> execBlock value. (testBlock value: blockValue) ifTrue:[^blockValue].
> Smalltalk garbageCollect. ^execBlock value.
>
I think its horrible personally but it /should/ work. The idea is that
files shouldn't need to be closed explicitly; which is fine; we want to be
able to rely on finalization to close them eventually. Hence retryWithGC:
is called when an open fails, presumably because it failed because the
process is out of file handles, presumably because several
as-yet-unfinalized files have yet to be finalized. So running the GC
/should/ run finalization. One issue could be that there are bugs meaning
that finalization does /not/ happen. For example, Igor could take a look
and find out quickly whether that's the case. Another issue could be that
the finalization loop is at too low a priority, or has been terminated for
some reason.
But finalization can not be relied upon to be timely. A better solution is
to explicity close files where possible using a form such
as forceNewFileNamed:do: et al. Another solution is possible in Spur where
we can simplify the finalization by using ephemerons so there are no
proxies for files. But that's not relevant immediately.
Huhhhhh, yeah, some lookup going on in a registry...
>
> and GCs? For file opens?
>
> I am doing these opens in a webapp, what I want is to know if the file is
> there or not and fail fast, not have all of this thing (debugging kind of
> nightmarish...)
>
> I have tested with 100 files and of course it all works since I get the
> IDs back from the registry.
>
> But at one point, it all fails. Registry full?
>
> What to do here?
>
> TIA
> Phil
>
>
>
>
>
--
best,
Eliot
