[bug#39647] [PATCH] gnu: Add unoconv.

[Tobias Geerinckx-Rice]

Jakub!

Jakub Kądziołka 写道：
> unoconv: format 
> [jw4yj39s84jb4fkvqpsqwq-unoconv-0.9.0/bin/.unoconv-real] is not 
> known to unoconv.
>
> I feel like this might be an issue with the packaging. Does it 
> work for
> you?

I don't see how this could be related to packaging…  And it works 
just fine on my laptop:

me@lapdog $ guix env --{pure,ad-hoc} unoconv coreutils¹ grep¹ \
           -- unoconv test-file.odt

me@lapdog $ # a pretty .pdf file is born!


But fine, I'll try it on another machine:

me@server $ guix env --{pure,ad-hoc} unoconv coreutils grep \
           -- unoconv test-file.odt
…


—and here's where I must stop to thank you for reporting the 
BESTEST BUG of 2020.  Seriously.  This made my day.  I laughed!  I 
cried!

Let's take another look at your error:

% unoconv ~/Downloads/scheme-macro-phd.ps -o scheme-macro-phd.pdf
unoconv: format 
[jw4yj39s84jb4fkvqpsqwq-unoconv-0.9.0/bin/.unoconv-real] is not 
known to unoconv.

So…  it's obvious that unoconv is erroneously parsing argv0 as 
target format.  Indeed, on both my machines, using the explicit 
‘-f <format>’

guix@env $ unoconv -f pdf test-file.odt


returns success and a .pdf file.

So is the long /gnu/store/… name overflowing a buffer?  Perhaps 
the wrapper script is to blame?  Maybe I should look at the co—

       ### If no format was specified, probe it or provide it
       if not self.format:
           ### Check if the command is in the form odt2pdf
           l = sys.argv[0].split('2')
           if len(l) == 2:
               self.format = l[1]


Oh.  Oh my.

I'd like to bet that the character before 
‘jw4yj39s84jb4fkvqpsqwq-unoconv-0.9.0/’ in your store name is a 
‘2’.  Because that's what it is on my server.  By sheer chance, 
the store name on my lapdog has TWO 2s, and so unoconv chugs 
blissfully onward.

I'm not sure what kind of fix upstream will accept for this.  At 
least a length limit, surely.  In the mean time it's trivial to 
just substitute* this out, since we won't ship any foo2bar 
symlinks our this package anyway.

> > +         (add-after 'unpack 'patch-find_offices
> I think that the convention is to replace underscores by dashes 
> in names
> like these.

Then I gladly thwart it.

> Wouldn't it be simpler to do
>
>     return [Office(...)]
>
> ?

Yep.

My reasoning at the time was literally: ‘well…  I should closely 
emulate upstream style [they use ret = … and append()] because 
I've never written a Python in my life, and they have, so they 
obviously know what they're doing.’

Needless to say, I'm currently reconsidering that opinion.

Thanks,

T G-R

[1]: ‘coreutils’ & ‘grep’ are undeclared dependencies of ‘soffice’ 
— will fix, but not relevant here.

signature.asc
Description: PGP signature