Spell checking identifiers

Dear c.compilers,


While experimenting with Rust, I came across this suggestion.


    --> foo.rs:5:9
      |
5 | return j; // the variable, not the type.
      | ^ help: a local variable with a similar name exists: `i`


Here it is suggesting i where I typed j. This is the same problem as
spell checking identifiers with fuzzy matching, so apologies for a po-
tentially misleading subject.


So, without going through the source of rustc to find out, I'm curious
about what general techniques people use to make this work? In particu-
lar the Damerau–Levenshtein distance algorithm is not appropriate for
dictionary lookups, as far as I know.


I've come across a survey of fuzzy matching algorithms, some of which
work with dictionaries but I have no idea which data structures would
be appropriate in a compiler, nor do I know what criteria I'd use to
choose an appropriate algorithm from such a survey.


As an added bonus, the same technique can of course be used to spell
check identifiers against a natural language dictionary. But since
such a dictionary is more static than the list of identifiers in the
current source file, a precomputed database will work, and a more
expensive indexing method can be used. Is there an indexing method
that works for this, but would not be appropriate for fuzzy matching
against identifiers?


[Apologies for not responding to my other topic yet, I should be able
to reply soon.]


--
Johann | email: invalid -> com | www.myrkraverk.com/blog/
I'm not from the Internet, I just work there. | twitter: @myrkraverk
[There's a vast amount of work on edit distance. My guess is they
use something like Levenshtein, but rather than use a constant
distance of 1 between different letters, the distance varies depending
on how different the letters look. -John]

Return to the comp.compilers page.
Search the comp.compilers archives again.