Re: Allow digits in identifiers (issue 6493072)

[dak]

On 2012/09/05 06:59:16, Keith wrote:
> While we are thinking about this, I suggest we remove (later) the rule
> forbidding backing-up states in the scanner.  It made only a 0.2% in
the
> worst-case scenario I could think of.

> The rule had been violated, giving us the slightly slower scanner, for
about ten
> years (I estimate) prior to the warning-cleanup patch da949cdc one
year ago.

> It costs a lot of programmer time to make the extra rules to save that
0.2%,

Not really.

> and
> I know of no efficient cure for patterns that can require arbitrary
amounts of
> backing up.  \violin1_2_3_4_5

I am not convinced that patterns requiring arbitrary amounts of backing
up make for a good choice of lexical units.

Whatever we allow as part of the language is something that convert-ly
needs to deal with, and every other system reading LilyPond code.
Including humans.  "Arbitrary amount of backing up" also applies to
them.

While I agree that requiring -b seems a bit arbitrary and is not an
absolute barrier and the performance impact will quite likely depend
heavily on the kind of backup patterns involved here, the use cases for
dropping them have not convinced me.

It also turns out that in the presence of large backup tables, it
becomes hard to guess which of multiple prospectively active rules will
get selected (have you tried looking at the respective debug output when
rules with backup states come into play non-trivially?).

So I am against removing this advice independently from patches that
might actually require this removal.

http://codereview.appspot.com/6493072/