I found the same thing. When I was writing Perl half the point was using regex. I inherited a site about 5 years ago that was a huge Perl site built entirely on text files and regex. The developer had basically built a database out of text files. He had directories for indexes, associations, etc.

It wasn't pretty to look at but it was still to this day the fastest site I've ever worked on.

How did he achieve transactionality?

Anywhere that he'd need transaction safety in the system, everything was contained in a single file.

There were other jobs that could rebuild indexes by scan, repair associations or break things down into parts of use on other areas of the site but the areas that needed safety were lumped together as a single record (similar to the NoSQL approach honestly).

It wasn't a general purpose setup, but it worked for his purposes.

Some elucidation on this point:

https://blog.codinghorror.com/regular-expressions-now-you-ha...

Great read. Thanks for sharing.

Regex patterns might not be first class citizens in python, but they're easy enough to use, and I think they're ultimately more powerful than in Perl. I don't remember being able to use a callback to calculate my replace string as a function of the match results when doing substitution in Perl. It's super easy to do tokenization with python regexes. We lose first class patterns, but gain first class functions and classes :)

> I don't remember being able to use a callback to calculate my replace string as a function of the match results when doing substitution in Perl.

That's easy with the `e` (eval) flag.

    say '13 37 123 42' =~ s
        { ( \d+ )           # get all the numbers }
        { my $total += $+;  # replace with running total }egrx;
    __END__
    output is 13 50 173 215

> I think they're ultimately more powerful than in Perl

No. Perl regex are more powerful than Python because

1. the programmer can pick from a large selection of delimiters

2. the full-featured engine is built-in and ready to use, but in Python you need to install the third-party "regex" module because the built-in one named "re" is hopelessly dyd¹

3. you may enable verbose (readable) character classes

4. the regex subsystem is pluggable with a common interface and you can within a lexical scope switch at run-time with more restricted implementations which have less features but better performance in some cases: https://metacpan.org/search?q=re%3A%3Aengine%3A%3A

5. you can embed arbitrary code with (?{...}) and (??{...})

> We lose first class patterns, but gain first class functions and classes :)

Perl had first-class functions since 1993.

You love having first-class classes in Python, but do not realise the downside. Python's one true meta object system (with its ageing design from the last millennium) is flawed: horizontal composition of methods with the same name does not emit a warning. Since the system is baked into the language, it cannot be changed. Contrast with Perl, where the language only provides some primitives on which to build a meta object system (incl. first-class classes). This enables bugs to be fixed, and the competition among several implementations on CPAN breeds excellence.

¹ https://i.imgur.com/66z4KDA.jpg

Not to mention that writing a composable first-class API to contruct regexes should be fairly elementary in Python.

oh what a good idea... wait, surely someone's already built this, right? like sqlalchemy's expression language, but for regex. If noone's built that yet, I'd build that.