Issues removing files with certain characters in their names.

[Harald Becker]

Hi Rich !

[To Rich: I have permanent mail failures on dalias at libc.org:
domain has no valid mail exchangers, answer is from GMX mail
server]


On 30-05-2014 14:12 Rich Felker <dalias at libc.org> wrote:

>ls never performas any filtering.

Sorry, yes. The name filtering is in the glob, not ls. My
mistake. So ls just displays the names from readdir. Beside that
mistake with filtering it's as I told.

>Printability has nothing to do with processing the filename. And
>a zero byte fundamentally cannot be in a filename (the filename
>in the directory table consists of those characters up to, and
>not including, any zero byte stored).

This is true on a Unix system, but have you looked what Windows
system do? Mapping of names is from to charset code pages is done
in so different ways, many of such mappings may produce unusual
characters. Those unusual characters can produce unusual
effects in further translations, ending up in illegal names when
translated to Unix file names.

With UTF-8 this is all getting better, but before name mangling
of foreign characters was a hell.


... you called it a user error when mounted incorrect. Yes, using
foreign characters in file names is the worst user error. All the
mentioned problems happen after a combination of translation
between different charsets. It is not a single point of failure
producing those problems. The chain of translations lead
sometimes to situations where file names from readdir let stat
fail with same name. All other name confusion can easily be
solved with pattern usage, but when stat fail with names from
readdir you are in trouble. Again, it should not happen, but it
happens - so call it shit.

And remember: All those translations are part of file system
drivers, not part of user space software. Most such userspace
software just passes names just straight through, without
modifications. So tell, who made the error?

--
Harald