bug#48902: 28.0.50; Directory names containing apostrophes and backticks cause problems

[Mattias Engdegård]

8 juni 2021 kl. 14.14 skrev Lars Ingebrigtsen <larsi@gnus.org>:

>> It's always possible that stringWithLispString isn't doing the right
>> thing. It's implemented at nsfns.m:3026. I know almost nothing about
>> UTF8/UTF16 so while it looks like it's doing the right thing to me, I
>> could be entirely wrong.
> 
> Right -- and that was written by Mattias, so I've added him to the CCs.

Thank you, but stringWithLispString: is actually fine, unless you count its 
inability to produce useful results from wrong input!

The image code seems to be quite confused with respect to whether the file 
names being passed around are in encoded form. Until recently it seems to have 
worked by pure chance since as long as the file name encoding is UTF-8 and the 
low-level code accesses the raw string data we do get the same result, but at 
least since 747a923b9a35 that's no longer the case.

Concretely, we have:

1. image_find_image_file probably expects its `file` argument to be 
non-encoded, but the string it returns is always encoded.

2. native_image_load calls image_find_image_file and passes its return value to 
ns_load_image.

3. ns_load_image calls [EmacsImage allocInitFromFile:] with its file argument.

4. [EmacsImage allocInitFromFile: file] can apparently be called with both a 
non-encoded or an encoded `file` argument (clearly not ideal), and it does:

  found = image_find_image_file (file);
  // This is dubious when `file` is already encoded.

  found = ENCODE_FILE (found);
  // This is completely useless since `found` is already encoded! Apparently 
ENCODE_FILE is idempotent, at least on macOS...

  [NSString stringWithLispString: found]
  // This produces nonsense as `found` is a string of raw bytes, so any Unicode 
will be converted to stretches of U+FFFD REPLACEMENT CHAR.
  [NSString stringWithLispString: file]
  // Same problem again, with a different variable.

The quick fix of reverting to stringWithUTF8String: will work, but the real 
problem is that we have no control of the encodedness of lisp strings being 
passed around. Comments would help, and I'd even go as far to suggest

typedef struct { Lisp_Object string; } encoded_file_name_t;

with the appropriate constructors and accessors, to get C's static type 
checking to work for us.