Option to use POSIX regex instead of PCRE about editorconfig-core-c

[ffes]

I am definitely not an expert when it comes to regular expression, but how does the syntax of POSIX regex differ from that of PRCE?

In other words, it this proposed change going to break anybody's .editorconfig file they have in their repo because a regex don't work anymore?

When I look at https://www.regular-expressions.info/refbasic.html and choose "PRCE2" and "POSIX BRE" I see for instance that a | is not supported in POSIX. Looking at other pages on the same site with the same flavors chosen I see a lot of differences.

[craigbarnes]

... but how does the syntax of POSIX regex differ from that of PRCE

It lacks a few features that are Perl-specific extensions, but the code in ec_glob.c shouldn't actually need anything besides the basic POSIX features.

In other words, it this proposed change going to break anybody's .editorconfig file they have in their repo because a regex don't work anymore?

Not if done correctly, no. .editorconfig files don't use regex patterns directly -- they use a custom format similar to the ones used by POSIX glob() and fnmatch().

From the git history, it seems like the code used to rely on an actual fnmatch() implementation (taken from FreeBSD), but the pattern format was later extended (beyond the capabilities of fnmatch()) and the code was changed to instead convert the patterns to regex (and then use PCRE for matching, purely as an implementation detail).

... and "POSIX BRE" I see for instance that a | is not supported in POSIX

POSIX BRE is considered a legacy format -- only kept in the spec for historical reasons. All modern POSIX platforms also support the extended format (ERE), which supports the | operator.

[ffes]

Thanks for the clarification 👍

I think one of the maintainers of this repo can answer your original question 😉

[xuhdev]

What is the application in your case that makes installing PCRE so burdensome?

[craigbarnes]

What is the application in your case that makes installing PCRE so burdensome?

My use case is static linking with musl libc. libpcre is about 400K of object code, which would triple the size of the binary I'm trying to add editorconfig support to, whereas musl's regex library is only about 20K.

Having an external library dependency also makes cross-compiling much more complicated vs. using a POSIX libc API.

[craigbarnes]

I've done this in a separate implementation, instead of trying to shoehorn it into editorconfig-core-c. I guess no one else really cares about using POSIX regex, so closing the issue.