This directory contains a Rust runner program for benchmarking the regress regex engine. The goal of this regex engine is to implement EcmaScript regex syntax and semantics.
The following decisions are made by this runner program:
- The
regressregex API doesn't permit searching anything other than a Rust&str, which implies the API only supports searching valid UTF-8. Therefore, this runner program always returns an error when the haystack is invalid UTF-8. - The
regressAPI does have a Unicode mode that can be toggled on or off, but it looks like it only currently impacts surface level syntax support.
Most regex engines support syntax like (?s:.) to enable flags for a
sub-expression of the regex. In the case of (?s:.), it enables "dot all"
mode in most regex engines such that . matches any character instead of its
typical default of matching any character except for newline terminators.
regress, likely because of its adherence to EcmaScript, does not support these inline flags. This means it cannot be used in some benchmarks that make use of these flags. In some cases, we try to rewrite the regex in order to avoid the use of these flags, but we don't expend the effort in every case.
Arguably, rebar could be expanded to push some of these flags into the
benchmark definition itself, similar to unicode and case-insensitive.
Although, that still doesn't resolve everything, because sometimes flags are
only enabled for parts of the regex and not all of it.
Similar to Go's regexp engine, regress has support for Unicode mode in some
aspects but not in others. For example, . and [^a] match entire codepoints
instead of individual bytes and case insensitivity is implemented by taking
Unicode case folding rules into account. Also like Go, though, the classes
\w, \d and \s are always limited to their ASCII definitions. Unlike
Go, as of regress 0.5.0, there is now a way of disabling Unicode mode. But
that Unicode mode seems to only impact surface level syntax, for example, by
rejecting invalid bounded repetition syntax.
One other thing worth mentioning is that regress has some support for Unicode
character class syntax, but tends to diverge from other regex engines in this
barometer. For example, regress supports \p{Letter} but not \pL, while
PCRE2, RE2 and Go's regexp package all support \pL but not \p{Letter}.
(It's likely that some or all of this support is dictated by ECMAScript, but I didn't spend the time to dig into that.)