Saving documents with broken zip streams (Re: minutes of ESC call ...)

[Luboš Luňák]

On Monday 03 of June 2019, Caolán McNamara wrote:
> On Mon, 2019-06-03 at 12:23 +0200, Luboš Luňák wrote:
> > On Thursday 30 of May 2019, Michael Meeks wrote:
> > >       + some in the zip area - assuming they are threading related.
> >
> >  Is this about those documents such
> > as /srv/crashtestdata/files/caolan/opendocument_stack_overflow_2.odt
> > ? How
> > can I reproduce that problem? If I try to fetch
> > buildslave at vm138.documentfoundation.org:
> > an/opendocument_stack_overflow_2.odt ,
> > it doesn't exist.
>
> I attach two of the examples here. The input name was foo.sample, the
> output to odt name appears higher up in the bt during the export.
>
> ./instdir/program/soffice.bin --headless --convert-to odt
> opendocument_stack_overflow.sample


 Ok, so it's not a problem with my code, my changes just happened to show the 
problem, and the problem is that those documents are broken. If you try to 
unzip the documents, it will complain about incorrect CRC (although it still 
will uncompress them). And what happens is that when we try to save the file, 
apparently only by that point we'll read those zip streams, there will be a 
ZipException about that, and the code in package/ is not exception-safe. So 
ZipOutputStream::writeLOC() gets called but not the matching 
ZipOutputStream::rawCloseEntry().

 But this is actually broken on several levels. If I make the code to catch 
the exception better, I'll need to make it somehow handle the fact that 
writeLOC() prepared for writing en entry, but then there's nothing to write. 
But that's actually not important, since ZipPackageStream::saveChild() will 
still return failure, so ZipPackageFolder::saveContents() will throw an 
exception, making the whole document saving fail. Which in turn means this 
whole save business is irrelevant, as there's just no way to save the 
document, even though we can load it and we can edit it. Which seems rather 
lame.

 Any idea what to do about that? Is it really ok that we just refuse to save 
it? Or should we save it even though the contents may be broken?

-- 
 Luboš Luňák
 l.lunak at collabora.com