Invalid variant ignored

[teg]

Using libkxbcommon (with the systemd-evcat test tool) I observe:

xkbcommon: DEBUG: Compiling from RMLVO: rules 'evdev', model 'pc105', layout 'us,us', variant 'dvorak-intl', options 'grp:caps_toggle'
xkbcommon: DEBUG: Compiling from KcCGST: keycodes 'evdev+aliases(qwerty)', types 'complete', compat 'complete', symbols 'pc+us+us:2+inet(evdev)+capslock(grouplock)'

(notice the missing trailing comma in the variant). The effect is that I get two plain us keymaps. I would have expected either an error message/failing compilation or for the compiler to somehow make sense of the missing comma (as X does AFAIU).

With the trailing comma:

xkbcommon: DEBUG: Compiling from RMLVO: rules 'evdev', model 'pc105', layout 'us,us', variant 'dvorak-intl,', options 'grp:caps_toggle'
xkbcommon: DEBUG: Compiling from KcCGST: keycodes 'evdev+aliases(qwerty)', types 'complete', compat 'complete', symbols 'pc+us(dvorak-intl)+us:2+inet(evdev)+capslock(grouplock)'

[bluetech]

This is the intended behavior; it is assumed to work this way by the "rules" file, so we can't change it really. From my tests X does the same, can you tell me how you tested it?

Coincidentally, since libxkbcommon 0.5.0 we now do print a warning when there is a user-provided RMLVO value which is not matched anywhere in the rules file (see d8a4f52). It turns out to not be very effective though because most settings are matched by wildcards whether they are valid or not (in the current rules file in xkeyboard-config at least). But it works well for misspelled "option" values, and it also warns in this case:

ran@ran:~/src/libxkbcommon$ ./test/rmlvo-to-kccgst -l us,us -v dvorak-intl
xkbcommon: ERROR: Unrecognized RMLVO variant "dvorak-intl" was ignored
keycodes: evdev+aliases(qwerty)
types:    complete
compat:   complete
symbols:  pc+us+us:2+inet(evdev)

Here are some technical details if you're interested:

I wrote a document a while ago describing what the rules file: https://raw.githubusercontent.com/xkbcommon/libxkbcommon/master/doc/rules-format.txt It is not "official", but I tried to follow the parser in libxkbfile exactly (used by setxkbmap and friends, and copied verbatim in the X server). In the Notes section I mention:

- If a Rule is matched, %-expansion is performed on the KccgstValue,
  as follows:

  %m, %l, %v:
     The model, layout or variant, if only one was given (e.g.
     %l for "us,il" is invalid).

  %l[1], %v[1]:
     Layout or variant for the specified group Index, if more than
     one was given (e.g. %l[1] for "us" is invalid).

And X has the same behavior:
http://cgit.freedesktop.org/xorg/lib/libxkbfile/tree/src/maprules.c?id=9fa04388dcbb5bf205319569f540a62c03754688#n545
It puts the "us,us" into indexes 1,2 of the layouts array, but puts "dvorak-intl" into index 0 of the variants array.

Since all the rules which match on both layout and variant look something like this:

! model       layout       variant       =      symbols

Or this:

! model       layout[4]      variant[4]     =      symbols

this doesn't match, it only matches these rules:

! model       layout[1]                    =      symbols
[...]
  *           *                    =      pc+%l[1]%(v[1])

And since v[1] doesn't exist, it is dropped, hence the variant is ignored.

[teg]

Thanks for the detailed explanation. I can confirm that an error is printed with 0.5.0, I thought I was testing with 0.5.0 all along, but as it turns out I was using the previous release (a package update had failed without me noticing):

$ ./systemd-evcat
You run in unmanaged mode without being root. This is likely to produce no output..
evcat - Read and catenate events from selected input devices
Running on seat 'seat0' in user-session '1'
Exit by pressing ^C or 'q'
xkbcommon: ERROR: Unrecognized RMLVO variant "dvorak-intl" was ignored

As to how X appears to have been working all this time with the wrong variant, I'm not sure. I assume something (Gnome?) may have been setting the correct layout for me and hence masking the error.