Regular expressions
Checking regex patterns at compile time, and where regular expressions go next
The shipped package nanyx.regex, with its #Matches and #MatchesRegex patterns for match arms, is described on
its reference page and in the cookbook, and how it is built, and
why, on Shipped packages. This page keeps
what is left: checking literal patterns at compile time, and the open questions.
Patterns checked at compile time
A bad pattern written as a literal should be an error where it’s written, not a crash when it runs. Nanyx doesn’t
evaluate code at compile time, so the check is to be a linter rule that calls the package’s own Regex.parse on each
literal passed to it, or to #Matches, and reports the RegexError at the column inside the literal. A pattern built
at run time still gives #error. No compiler special case is needed.
This waits for linters written in Nanyx (#107,
#1343): the linters are F#, compiled into the
compiler, and can’t run Nanyx code. Porting the pattern reader to F# would make a second parser that drifts from the
package’s, and compiling and running the package from the F# linter would be machinery thrown away once the contract
exists. Until then, a wrong literal gives #error from Regex.parse, and #Matches crashes with it.
Open questions
- Parser combinators. A
nanyx.parsepackage is often the better tool for structured text, and reads well with pipes and workflows. It would complement regexes, not replace them. - Where it lives if the standard library’s public modules are reorganised
(#2029): it fits under
nanyx/textif it moves into the standard library.
Plan
| Step | Issue | After |
|---|---|---|
| A linter rule that checks literal patterns at compile time | #2033 | #107, #1343 |