Why does Windows have trouble finding my Win32 resource if it contains an accented character?

Maurice Kayser reported an issue with Win32 API loading of PE resources containing lowercase letters. Maurice did some experiments adding resources named MyIcon, MyIcÖn, and MyIcön, then trying to load them using various names, and built up a table of results. I’ve broken it up into three tables depending on the nature of the accented character.

This table shouldn’t be surprising. The argument passed to LoadResource is compared case-insensitively with the name of the resource, treating accented characters as different from their unaccented versions.

Here’s the next batch.

The first column is consistent with our previous result, namely that unaccented characters are treated as not the same as accented characters.

The second column is not surprising either, since the strings do match according to a case-insensitive comparison.

The third column is surprising. It seems that accented characters are case-sensitive, even though the documentation says that the comparison is case-insensitive.

Okay, here’s the third block.

The PE specification says that the resources are sorted “in ascending order”, and the names are sorted “by case-sensitive string.”¹

That’s all it says. The rest is left to interpretation.

First of all, even though the file format specification says that the resource names can be in any case, the FindResource function converts all names to uppercase before searching, so any names with lowercase characters are effectively unfindable. Fortunately, the Resource Compiler also converts names to uppercase before storing them in the resources, so it all cancels out, right?

Well, it cancels out only if the Resource Compiler and the FindResource function agree on how the names are converted to uppercase.

The Resource Compiler uses _wcsupr to convert the names to uppercase, and _wcsupr uses the default C locale,² which as we noted before, is not a very interesting locale. It converts Latin unaccented lowercase letters a-z to Latin unaccented uppercase letters A-Z, and that’s all.

Let’s update the top row of the table by converting the names to uppercase according to the C locale.

How does the FindResource function convert strings to uppercase? It uses the uppercase table corresponding to the system default language. It is almost certain that Ö and ö are uppercase and lowercase partners in the system default language. That means that the left columns are all effectively MYICON in the first table, and that they are all effectively MYICÖN in the second and third tables.

With these adjustments, the tables make more sense.