Re: random maddening bug - "corrupted" file upload contents

Cyrille,

I wrote a file like this:

FileLocator temp / 'myzip.zip' writeStreamDo: [ :out | 
  out binary; nextPutAll: #[80 75 3 4 10 0 0 0 0 0 125 83 67 73 0 0 0 0 0 0] ].

Which I can read as follows:

FileLocator temp / 'myzip.zip' binaryReadStreamDo: [ :in | in upToEnd ].

  => #[80 75 3 4 10 0 0 0 0 0 125 83 67 73 0 0 0 0 0 0]

If I upload this file to http://localhost:1701/form-test-3 this works perfectly (macOS 10.13.2 Pharo #60402, Safari. Firefox & Chrome), first time and second time.



Maybe you should try on a co-workers machine just to make sure it is not a local problem on your machine ?

Sven

> On 22 Dec 2017, at 15:44, Cyrille Delaunay <[email protected]> wrote:
> 
> Thank you sven for the advice.
> I think have been able to reproduce the issue by only playing with  
> 
> http://localhost:1701/form-test-3
> 
> First of all, here are several ways I can retrieve the expected bytes constituting my file:
> - From terminal: 
> "od -t x1 myfile.zip "
> => gives me:  50  4b  03  04  0a  00  00  00  00  00  7d  53  43  49  00  00 ...
> (which matches in decimal:  80 75 3 4 10 0 0 0 0 0 125 83 67 73 ...  )
> 
> - From Pharo:
> 'myfile.zip' asFileReference binaryReadStreamDo: [ :in | encoding := (ZnCharacterEncoder detectEncoding: in upToEnd) ].
> (('myfile.zip'  asFileReference binaryReadStreamDo:
>             [ :in | (ZnCharacterReadStream on: in encoding:  encoding) upToEnd ]) copyFrom: 1 to: 40) asOrderedCollection collect: #codePoint"
> => gives me:  an OrderedCollection(80 75 3 4 10 0 0 0 0 0 125 83 67 73 0 0 0 0 0 0)
> 
> 
> Now, after triggering the upload of this file from  http://localhost:1701/form-test-3, 
> I have been able to halt in method #executeOneRequestResponseOn:,
> and to notice again something that looks wrong regarding the first four bytes:
> 
> <Screen Shot 2017-12-22 at 9.48.17 AM.png>
> 
> The web interface displayed something confirming this "strange" first bytes
> 
> <Screen Shot 2017-12-22 at 3.32.19 PM.png>
> 
> 
>  Right after, I did the exact same upload test again and halt at the same location.
> And this time:
> 
> <Screen Shot 2017-12-22 at 3.35.43 PM.png>
> 
> 
> <Screen Shot 2017-12-22 at 3.38.39 PM.png>
> 
> 
> Problem is gone !
> I am starting to pull out my hair from my head :)
> 
> 2017-12-22 11:54 GMT+01:00 Sven Van Caekenberghe <[email protected]>:
> Cyril,
> 
> If you think something is wrong with the low level Zinc implementation, you could try without Seaside.
> 
> ZnServer startDefaultOn: 1701.
> 
> http://localhost:1701/form-test-3
> 
> You could try to debug/inspect from #formTest3:
> 
> HTH,
> 
> Sven
> 
> > On 22 Dec 2017, at 11:38, Cyrille Delaunay <[email protected]> wrote:
> >
> > Hi everyone,
> >
> > Since some times, we are experimenting a random bug while using seaside file upload.
> > Not sure at all about the exact origin of this issue (Application specific, Zinc, Seaside, Pharo ?).
> > But as it start to drive us mad, I'm sharing our problem with you.
> > Maybe someone would already have experimented something similar or would have a good feeling about the issue.
> >
> > High level description:
> >
> > We have a file upload element rendered with seaside.
> > The callback of this file upload element is triggered when the form is submitted via a submit button.
> > Problem: Within this callback, the raw contents of the WAFile does not match the expected contents of the selected file.
> > The first 4 bytes are "wrong", and rest of the byte array seems ok.
> > This issue occurs randomly and we haven't been able to determine / guess any context criteria.
> >
> >
> > Debugging:
> >
> > In order to figure out what was wrong, we tried to catch a place in pharo where file contents started to differ from the expected file contents.
> >
> > The earlier upload step in which we have been able to notice this delta, is method
> >
> > ZnMultiThreadedServer >> executeOneRequestResponseOn:
> >
> > In this method, a http request is read and materialized from an input stream,
> > From my naive point of vue, this stream looks like a kind of communication entry point between the web browser and pharo.
> >
> > The zip input file I am trying to upload is expected to begin with the following bytes:
> > #[80 75 3 4 10 0 0 0 0 0 125 83 67 73 ...
> >
> > The request object built within #executeOneRequestResponseOn:  points to an"entity" variable, which points to a "parts" variable, which points to a single ZnMimePart.
> > This ZnMimePart points to a byte array beginning with:
> > #[0 0 0 0 10 0 0 0 0 0 125 83 67 73 ...
> >
> > I also experimented another case where
> > The request object built within #executeOneRequestResponseOn:  points to an"entity" variable, which points to a "parts" variable, which points to multiple ZnMimePart.
> > One of the ZnMimePart points to a byte array beginning with:
> > #[138 67 80 75 10 0 0 0 0 0 125 83 67 73
> >
> > We can see that in both cases the first 4 bytes do not match the expected first 4 bytes of the file.
> > As a consequence, later in our process,
> > pharo gets a error while trying to unzip this upload file, complaining about a "'bad LH signature at 0'".
> >
> >
> > Does this speak to anyone ?
> >
> >
> >
> > --
> > Cyrille Delaunay
> > _______________________________________________
> > seaside mailing list
> > [email protected]
> > http://lists.squeakfoundation.org/cgi-bin/mailman/listinfo/seaside
> 
> _______________________________________________
> seaside mailing list
> [email protected]
> http://lists.squeakfoundation.org/cgi-bin/mailman/listinfo/seaside
> 
> 
> 
> -- 
> Cyrille Delaunay
> _______________________________________________
> seaside mailing list
> [email protected]
> http://lists.squeakfoundation.org/cgi-bin/mailman/listinfo/seaside

_______________________________________________
seaside mailing list
[email protected]
http://lists.squeakfoundation.org/cgi-bin/mailman/listinfo/seaside