flex-2.5.31 [was: environmental issues]

[David Relson]

At 05:29 AM 8/4/03, Matthias Andree wrote:

...[snip]...

> > Flex 2.5.31
> >       However, it seems that pattern "<INITIAL>^[ \b\t]*\n" is
> >       evaluated at a different point in time, which results in
> >       bogofilter treating the first line of the message body as part of
> >       the message's header.
>
>Is there any ambiguity in the flex patterns? If so, we should fix that.

That's not the issue, AFAICT.  The "%never-interactive" option works 
differently in flex-2.5.31 than in flex-2.5.4.

Using flex-2.5.4, with "%never-interactive" set, the transition from 
message header to body works properly.  Using flex-2.5.31, the pattern for 
an empty line is matched later and the header/body transition is wrong, 
causing bogofilter to fail.  With flex-2.5.31, removing the 
"%never-interactive" line causes the lexer to work like flex-2.5.4 does.

Attached is test.flex-2531.tgz which contains the test code and data:

lexer.l - a near minimal version of bogolexer.
msg     - a near minimal email
tst     - a script to build two versions of lexer.l, run them, and compare 
the outputs.
         - assumes flex-254 is the flex-2.5.4 executable and flex-2531 is 
flex-2.5.31