Some attached files with non-ASCII names are unlinkable

[filosofo]

This will be tough to reproduce in current trunk while the inline uploader is missing, but you can see it happen in 2.3.2.

1. Upload a file named WordPress › Error.jpg
2. wp_unique_filename, calling sanitize_title_with_dashes, which in turn calls utf8_uri_encode, changes the filename and its url to wordpress-%e2%80%ba-error.jpg
3. When WordPress links to this file as an attachment, it breaks the link. That's because the server sees the encoded text %e2%80%ba, and converting it back, thinks you're requesting wordpress-›-error.jpg, a file that doesn't exist. To be served a file named wordpress-%e2%80%ba-error.jpg, one would have to request instead wordpress-%25e2%2580%25ba-error.jpg

My patch strips out the percentage character (%) in wp_unique_filename, so that this confusion is avoided. It seems somewhat like a hack, so perhaps someone else can think of a better solution.

[filosofo]

duplicated in #5826

[hakre]

It would be great to see this inside the next revision (2.5.1). In the support forums many users report errors with the new uploader and this reduce some of them. I just checked against version 2.5 and 2.6 bleeding and the error is still in. This patch would at least fix the problem. Therefore some naming information from the filename is ok to loose.

[hakre]

Additionally, a comment should be put inside the sourcecode to reference this ticket.

[ryan]

(In [7630]) Strip percent signs when sanitizing filenames so the server doesn't try to decode entities. Props filosofo. fixes #5587 for 2.5

[ryan]

(In [7631]) Strip percent signs when sanitizing filenames so the server doesn't try to decode entities. Props filosofo. fixes #5587 for trunk