Rockbox

The problem seems to be that 'genlang' takes the last pattern that matches for each phrase, so using "*:" before "recorder:" will work.

This begs the question: should we change this in genlang or adjust the lang to work like this?

Thanks for investigating Bagder. However when switching around
the strings so that the "*:" is before "recorder:" I get the exact
same behaviour –> The strings appear in the wrong places.

I attached a screenshot of the main menu, it looks the same both
with the above example and with this:

<phrase>

id: LANG_CONFIRM_SHUTDOWN
desc: in shutdown screen
user:
<source>
  *: none
  recorder: "Press OFF to shut down"
</source>
<dest>
  *: none
  recorder: "Press OFF to shut down"
</dest>
<voice>
  *: none
  recorder: ""
</voice>

</phrase>

   mainmenu.png (2.1 KiB)

Well, I must admit I didn't do a full build with my work but I'm trying to sort out the issue on single file level first. I've attached my "test.lang" that I used to repeat the problem and work with this issue.

With the test.lang file, I generate C and H files using a command line like:

(for h120)
$ genlang -e=test.lang -p=testlang -t=h120 test.lang

(for recorder)
$ genlang -e=test.lang -p=testlang -t=recorder test.lang

Both these command lines generate testlang.[ch] and they seem correct for me when I use the order as in the attached file.

Ehum. Are you saying that the built-in version works fine for you but that the .lng file is broken?

Doing the same language but generating a .lng file is done like this:

(for h120)
$ genlang -e=test.lang -b=testlang.lng -t=h120 test.lang

(for recorder)
$ genlang -e=test.lang -b=testlang.lng -t=recorder test.lang

… and they seem to differ for that shutdown message.

Can you see any obvious errors when using the tool like this? I'm not saying there's no error, I'm just trying to understand where it appears…

   test.lang (0.8 KiB)

Bagder, thanks again for looking in to this! Yes, my problem is that the built in strings are correct but in the .lng files they are not.
(I'm sorry if I was unclear)
I now made a test with your test file and it has the same strange behaviour that I saw in
englis.lang with the example above.

$ genlang -e=test.lang -b=testlang.lng -t=h120 test.lang
with your test.lang produces a 21 byte testlang.lng

when I completely removed the phrase for LANG_CONFIRM_SHUTDOWN and ran
$ genlang -e=test.lang -b=testlang2.lng -t=h120 test.lang
command I once again got a 21 byte .lng file.

"$ diff testlang.lng testlang2.lng
Binary files testlang.lng and testlang2.lng differ"

I opened them in a hexeditor and found the one place they are different.
The byte before "Maybe" is "03" in testlang.lng but "02" in testlang2.lng
I assume this is the string ID number which makes the numbers in the testlang.lng
file go like this: 00Yes..01No..03Maybe

and in testlang2.lng (where the recorder string was removed from the .lang)
00Yes..01No..02Maybe

So it seems the ID numbers are wrong.

Thanks for your details.

The culprit of this problem is the code starting at genlang line 175 that deals with reading an english .lang file to get the english order to use for output. This code does not deal with 'none' so it will count all IDs present in the file. The same code is not used when the C and H files are generated so for this case, 'none' will completely removed phrases marked as 'none'.

The only sensible fix here is to fix the code that gets the english order to properly check for and acknowledge the 'none' keyword. It isn't however a trivial fix but will take some further parsing and calling the pattern match function etc. I'll try to get some time for this soon if nonody else does it.

A fix for this is now in SVN. Can you see if things now work for you?

Genlang now generates a correct test.lng from the test.lang file posted above so it does indeed look good.
I hope to get the langv2 cleanup in sync tomorrow and test with that too.

Good news, this does indeed fix the problem above (although with a small quirk, it only
works for 'none' if I write 'NONE' it behaves as before, with gaps in the id numbers, took
me a while to find out but now it doesn't matter much)

Bad news, another problem surfaced that creates almost the same effect buildzip.pl does not send the features from features.txt to genlang when building the
.lng files it only sends the regular target name, so any strings that only have a feature
match and no match to the target name will be excluded…

If I find some time I will try to find a solution but I have so far in my life never written
a single line of pearl, so it will probably take a while…

Anyway thanks for the fix, one step cloesr to getting the cleanup comitted.

Ah, yes. 1) The NONE issue can probably be fixed by switching to proper case insensitive checking all over. 2) The buildzip problem will need a fix similar to what is used in the build makefile for the features.

I'll try to have a go at these soon.

Unfortunately, genlang in current svn doesn't handle -u well at all at the moment. Basically, -e doesn't read anything. Using 'dest' to parse the string in this case is incorrect, as a target hasn't been specified. Specifying a target does make things work better.

Also, there seems to be some debug output left.