Hi Adrian, > Le 19 janv. 2020 à 19:41, Adrian Vogelsgesang <[email protected]> a > écrit : > > Hi, > > As reported by Hans Åberg in > https://lists.gnu.org/archive/html/bug-bison/2020-01/msg00021.html, the C++ > skeleton lalr1.cc accepted arbitrary values for parse.lac.
Yes. Too bad it was caught after the release of 3.5.1, I would have wanted it to be part of it. > The attached patch makes sure we only accept the two values “none” and “full” > and reject all other values. You can also find this patch on > https://github.com/akimd/bison/pull/23 Great! > I am not completely convinced this patch is the right solution, though. > Although not intentionally, version 3.5.0 and 3.5.1 accepted “-Dparse.lac” to > enable LAC-support. With this patch, “-Dparse.lac” is rejected again and > “-Dparse.lac=full” has to be used instead. Personally, I think we can live > with this breaking behavior, In particular given that “-Dparse.lac” was never > documented and I would prefer to keep the set of supported parameters as > small as possible. > What do you think? I would have asked myself exactly the same questions, reaching the same conclusion. I'll wrap 3.5.2 with this patch in a week or two, just in case something else pops up. Maybe I'll be done with the CI under Windows, and more tweaks might be needed. Thanks for the prompt reaction. A few minor issues: > Just as the yacc.c skeleton, the lalr1.cc skeleton should reject > reject invalid value for parse.lac. Too many "rejects", and I guess "values". +AT_BISON_CHECK([[-Dparse.lac=none input.y]], [[0]], [], []) +AT_BISON_CHECK([[-Dparse.lac=none -L c++ input.y]], [[0]], [], []) +AT_BISON_CHECK([[-Dparse.lac=full input.y]], [[0]], [], []) +AT_BISON_CHECK([[-Dparse.lac=full -L c++ input.y]], [[0]], [], []) Rely on the default values here, pass only the first argument. Cheers!
