Inconsistent performance for gencache when there is same filename with different suffix.

[lychees]

Early today, we found that gencache might have some encoding error when executed in Linux server.

After some research, we think this is a problem with the behavior of gencache.

index.json generated in web servers with linux.

indes.json generated in local PC.

The resource files.

I think gencache generated different files in different OS. (Afterall, there might even be some case issues under Linux.)

[lychees]

Since we are using small case file name without suffix as map key, I thought there might be some bug, if there is some files with same file name but different capitalization and suffix name? I am not sure.

[Ghabry]

So you are having two files next to each other. One with extension and one without, e.g.

• CharSet/Image.png and
• CharSet/Image ?

The behaviour is in this case indeed undefined. The result depends on the enumeration order of the filesystem. Windows always reports in alphabetical order and Linux in order when a file was created.

This will be even a problem for our OpenRTP where I think about putting license files next to the image with txt ending.

[lychees]

Another case for translation files, it might cause some system cannot detect those i18n files.

[Ghabry]

this is probably caused by an out-of-date gencache. Download the latest version. Not removing "po" or "ini" is a change made 2 months ago.