In my opinion, those are results of Perl losing mindshare. Numpy is not exactly hard (R has a significant fraction built in, and there's an old standard library for Scheme, of all things, that is very similar). Perl was perhaps the leading language for websites, but the Perl 5/6 situation just confused everybody as it dragged on, which made Python and Ruby libraries seem more attractive.

> And I guess Perl also never had its Django (or Rails).

It’s got two of them - Catalyst (older still works fine) and Mojo - newer and shinier.

I think python is a better fit for mathematical work. I like to say that python helps you think more like the computer does, and that perl helps the computer think more like you.

Catalyst was no match for Rails. Mojolicious could have been had Miyagawa's PSGI/Plack appeared on the scene half a decade earlier.

rails as far as I understand is opinionated and optimised for CRUD / database type applications. Catalyst is much more agnostic about the model you use. It provides the flexibility and a way of structuring the code and providing debug tooling that makes structuring a decent size app well reasonably easy to do.

Catalyst was painful to work with in my experience. Just getting it setup was an achievement. Subroutine signatures as routes is an ugly hack.

I like catalyst,s dispatcher. And some of the pain of installing catalyst in the early days resulted in huge improvements to the change toolchain.

I honestly think the Perl community's attachment to Catalyst contributed to Perl5's demise. Mojolicious was a much better bet for the future of Perl.

I've only done a little bit of mojolicious, but it's quite nice. The Mojo::UserAgent / Mojo::Dom tools are certainly great for writing web scrapers.