Incompatibility between Bison 1.875 and Flex 2.5.25

[Paul Eggert]

> From: Akim Demaille <address@hidden>
> Date: Mon, 06 Jan 2003 09:44:20 +0100
>
> YYLTYPE was a name invented in the Bison world: how can they use it
> without asking us?

I think flex is trying to generate a file that can be used along with
a Bison-generated file.  That is, flex is not defining YYLTYPE; it is
merely using "#ifdef YYLTYPE" and assuming that if the type is defined
as a macro, then Bison is generating line number info.  This
assumption was incorrect for Bison 1.75b, as it defined YYLTYPE as a
macro   even when it was not generating line number information.

But hmm, come to think of it, the assumption is also incorrect for
Bison 1.875.  In Bison 1.875, YYLTYPE is a typedef, not a macro: this
is for consistency with YYSTYPE, which POSIX requires to be a typedef.
This change first appeared in Bison 1.75e.

I guess we have to choose between consistency with POSIX and
compatibility with flex 2.5.23 (dated 2002-10-21) through flex 2.5.25
(dated 2002-12-02).

I didn't know that Flex test versions were available.  I'll CC: this
message to W. L. Estes, since he is generating those versions.
Perhaps he can suggest a way of fixing flex so that it works with the
new Bison regime, or vice versa.  Bison 1.875 does define a macro
YYLTYPE_IS_DECLARED which flex might want to use instead.  Or perhaps
we should define a more standard (and documented) macro regime for C
code that needs to know which Bison options were used, and then have
flex use that.