All PsdBooleans are True

[widdowson]

Hi there Christoph,

psdtags is great! But I just found a bug while doing an overly[*] paranoid assert(originalbytes == frombytes(originalbytes).tobytes() at the PsdLayer level). It seemed to me that a layer's infx value should have been 0000, but it was being written out as 1000. The same thing was true of knko. But I noticed that there wasn't an unexpected diff in the layer's clbl. They're all PsdBooleans. Then it occurred to me.. it's because they were all hardcoded by psdtags to 1000 (True), and only the clbl was originally true. Lo and behold:

In https://github.com/cgohlke/psdtags/blob/v2024.1.15/psdtags/psdtags.py#L2490 we see:

        value = bool(fh.read(1))

However:

% python3
>>> bool(b'\x01')
True
>>> bool(b'\x00')
True

I believe this is is making all PsdBoolean's read()/frombytes() True, regardless of input value. This also isn't the only place I see a bool(fh.read(1)), which I think is generally a bug for the above reason - unless the read literally returns False or None, you're always going to be True.

I tried patching this in my local checkout to check for fh.read(1) != '\x00', but it had really bizarre side effects - namely that write()/tobytes() output of my PsdLayer object ballooned in size. If I just hardcoded value = True I had identical broken output, but if I got it to parse correctly, I think some other routine is conditioning on some PsdBoolean somewhere, and since it's never been False before (or at least recently - I didn't check blame history) I'm guessing said codepath wasn't exercised.

Hope this helps,
-Andrew

[*] turns out maybe this wasn't overly paranoid, but "just right paranoid" ;)

[cgohlke]

Thanks for reporting. bool(fh.read(1)) is definitely wrong, should be bool(fh.read(1)[0]).

I rely on the assert to catch cases like this. So far all the files on my system passed with that bug. I certainly don't have enough files for testing and don't have easy access to Photoshop to create new files...

[cgohlke]

I can't seem to reproduce the bizarre side effect, ballooning. Can you share a file and/or code?

[widdowson]

Happy to help report bugs, always :). I think there's something going on with my use of channels in PsdLayers that might be causing the problem, and I don't want to waste your time on that since I'm now more convinced it's a glitch on my side. Whoops!

[cgohlke]

The PsdBoolean read issue should be fixed in v2024.2.22.