Re: Help with debugging apparent error in parser

[Trevor Daniels]

Carl

You may be right, but the details of the parser/lexer logic
have so far defeated me.  As it's late and as I'm away most of
Saturday I shall withdraw, beaten, from this discussion :(

Trevor

----- Original Message ----- 
From: "Carl D. Sorensen" <address@hidden>
To: "Trevor Daniels" <address@hidden>; "lily-devel" 
<address@hidden>; <address@hidden>
Sent: Friday, August 08, 2008 10:10 PM
Subject: Re: Help with debugging apparent error in parser




On 8/8/08 3:09 AM, "Trevor Daniels" <address@hidden> wrote:

> Carl
>
> This isn't a problem specific to chordmode.  If you use
> an identifier c you get a similar error about unexpected
> notename_pitch whatever the input mode.  The problem is
> the identifier m.
>
> Because m is a valid chord modifier and the parser/lexer
> is scanning for chordmodifiers or pitchnames it considers
> the m to be a chord modifier rather than a variable name.
>
> It seem the parser/lexer relies on finding a token which
> is invalid in the current input mode to cause termination
> of that mode.  So your first expression is ok because m
> is invalid in notemode.

I would agree with this, except for the problem that I can do

\myDisplayMusic \chordmode{c}
\myDisplayMusic \notemode {c}

which should put the parser back in notemode, and I still get the error with
m as an identifier.  However, if I write

\myDisplayMusic \notemode{c}

I do not get the error with m as an identifier.  Thus, it appears to me that
once the parser enters chordmode, it sets something that it doesn't clean up
when it moves from chordmode to notemode.


> Such errors can be avoided by:
>
> (a) placing all variable definitions before music

There is some justification in the documentation for being able to use void
music functions to set parser behavior in multiple locations.

> (b) not using single-character variable names.

I agree.  I don't use single-character variable names.  This particular case
was identified in a regression test.

I could change the variable name in the regression test.  However, there is
still a subtle bug that I have been observing that has to do with the parser
entering chordmode in a void music function that causes strange errors.

This experience causes me to believe that there is a bug in the parser
related to chordmode.

Thanks,

Carl