Hacker News Reader

Comment by proverbialbunny | Hacker News Reader

I loved Perl back in at the turn of the millennia. My first data science equivalent projects were in Perl.

To this day I can not find anything quicker to prototype in. Python is slow in comparison. Perl is great for quickly hacking something together for R&D purposes. Also, if you need to write something custom, Perl runs so much faster than Python and R.

I, like many people, stopped using Perl beyond a quick script in 2010. Not out of a lack of love. Perl was great. I wonder if your experience comes from being around the laggards, not so much Perl devs itself. When I was working in Perl no one around me was afraid of Javascript. Javascript and Perl are both C based. They aren't that different.

Your story is eye opening. It tells me what it would have been like if I had stuck around. I imagine working with the few who didn't want to learn new skills must have been more painful than you're letting on.

View on HN

I wrote a lot of popular web apps in Perl, some that appeared in NY Times and Time mag. I quit around 2010 and went over to php which is easier. Wrote a great app and built a successful software biz around it. I think 2010 may have been when Perl 6 felt hopeless and there was no clear leadership or direction in the community.

Mod_perl was the sh*t back in the late 90s for super high performance web apps on Apache. Then Perl stalled and everything else didn’t.

Perl could have been where Python is today if they didn’t take 20 years to waffle about a release that was supposed to do everything and ended up doing nothing because everyone left.

What's also forgotten is the Perl community's endless procrastination in adopting a meta-object protocol (MOP) and the too-late arrival of Miyagawa's brilliant contribution - PSGI/Plack. These 2 factors, I believe, led to Perl being overtaken by Rails, Django and the various PHP frameworks.

Perl’s lunch was eaten more by Ruby IMO. The bifurcation happened as the Perl programmers that wanted a more traditional OOP system picked up Ruby and the folks writing Bash++ oftentimes went to Python. These days it seems like Ruby is where Perl was in 2010.

    These days it seems like Ruby is where Perl was in 2010

It's more "mature" than "abandoned" IMO. I think it's in a solid place. A lot of Rubyists have moved on but so far, that's OK with me.

Unlike OP's experience with Perl, most legacy Ruby and/or Rails codebases are not too disasterous.

That's probably because Rails hits the sweet spot. It's just complex enough to support a medium-sized website well. Most innovations since Rails have targeted high-load websites.

mod_perl was dropped with the release of apache 2.4 in 2012 (that's a major release, apache is not semver). If I remember well, it was kicked out of the apache repo.

mod_python was also abandoned around the same time. It simply stopped being developed and stopped working (it's crashing on python 2.7).

Basically applications stop running inside apache memory with mod_xxx for good reasons (it's unstable and crashing both apache and the app). Languages moved to a FastCGI interface or to have a standalone web server.

Do you mean faster to write, or faster to run, or the combination of both?

Both I would guess.

Perl in serial is already faster to run than Python, and as soon as you start using threads Perl blows Python straight out of the water. The only reason that stuff like tensorflow or numpy is not as dog slow as the rest of python is that they are bindings to c(++) or fortran libs.

Writing faster is debateable, but Perl has tons of niceties like embedded regex syntax, string interpolation, one-line if/unless that make Perl both more concise and less boilerplatey, which arguably also makes it faster to write.

There is also "faster to maintain". Perls compatibility means that there is almost none of the churn e.g. Python has with regard to syntax and semantics changes. Perl code from almost 3 decades ago usually works quite fine, whereas Python code breaks in every minor version upgrade and needs a complete rewrite for major versions

I've been writing Python for a good while now and I haven't seen too many issues between minor versions.

As for major versions...well...there's only been one significant change in this area (2 to 3) so I feel like maybe you're hyperbolizing a little :-)

Python 3 was released in 2008, Python 2.7 had some kind of release last year but my memory is that up to about 2013 there was still a serious question of 2.7 vs 3 when you started a new project ("I know we should use 3, but then we can't have x-lib!").

So it happened once but it took up about 25% of the last 20 years.

A little. But there were changes from e.g. 2.5 to 2.7 and in the earlier 3.x versions that broke stuff. That never happened with Perl code here.

>Perl in serial is already faster to run than Python

Not it isn't.

>The only reason that stuff like tensorflow or numpy is not as dog slow as the rest of python is that they are bindings to c(++) or fortran libs.

So if I make a foo.py file and run the code, it runs faster. "but it's really c(++) or fortran!" doesn't matter in the slightest.

Both. Perl is around 20x slower than C for common tasks, while Python is around 133x slower than C for common tasks like looping, depending on what you're doing of course.

Perl is the ultimate mockup language in that you can crank out code very fast and get good bug free results. It lets you write in the way you think where other programming languages force you to write a specific way. Ofc you can get good at Python, so good at it you might get as fast as Perl. And Python is easier to read. The advantage of Perl, writing like you think, makes it difficult for anyone but you (or someone else who thinks like you) to read and maintain your code. The ugly code stereotype comes into play from this.

Perl isn't like Python. It's great if you want to write a one off script for doing quick analytics or a one off script that checks something on a server. It's fantastic for that. Python, you'd need to google around for a library, hope you find one, grab it, write a 100 lined file, and then finally be done, where Perl was a quick one-liner to do the same thing. Ofc the Perl is going to look like garbage and not be maintainable and the Python is going to look golden and be maintainable. They both fit a different need and different use case.

> Ofc the Perl is going to look like garbage and not be maintainable

Please stop perpetuating this myth. It really depends how you write it. Perl isn't going to do the tidying up for you, you'll have to properly structure the code yourself. If you still write Perl code like it’s 1991 then maybe itcs not maintainable. I work on a largeish Perl code base and I can assure you it's quite maintainable, moreso than most Js that people install via npm as we speak.

> If you still write Perl code like it’s 1991 then maybe itcs not maintainable.

This is probably one of the issues with Perl. You can still write Python code like it's 1991 and it will still be readable and maintainable.

no you cant, cuz 2.x python wont work in 3.x ;)

Fair enough, but note that the discussion topic is "writing xyz code like it's 1991", which doesn't exclude using a modern compiler.

I don't think your performance numbers are very accurate. I've done a bit of Perl and a lot of Python and they're fairly comparable in most cases. Python is lightning fast for a lot of scripting tasks and I generally use very few libraries for day to day tasks because the Python stdlib is so good.

Unless you're doing something really simple with Perl command line switches that operate over an entire file, I'd bet Perl and Python solutions will be in the neighborhood of the same number of loc even with all the magic in Perl as Python has a lot of modern conveniences. It's easy to build a complex datastructure in Python with nested dictionaries with sets/lists/other dictionaries... whatever. In Perl it's more complex.

No, his estimates look pretty accurate to me. I've also done a fair bit of both Perl and Python (including working with the internals in C).

Perl is significantly faster than Python for many types of common string manipulation tasks. They're both fast, sure, but Perl has some key optimizations.

For maybe some regex tasks sure, but there are also plenty of Python operations that are significantly faster than Perl, not to even mention all the data science stuff. I might have misunderstood the comment as evaluating the language on one thing doesn't capture a broad/general spectrum of usefulness.

Every language in this category will have bindings to domain specific high performance modules. This is part of why performance doesn't matter all that much - we can always push hot operations into another language.

Perl generally outperforms python in core language features.

The benchmarks shown elsewhere here tend to disagree with that. I think at one point that was more true than it is now. Overall though they seem to be comparable enough to where it probably doesn't matter.

Also, the bindings have to exist which they don't with Perl for a lot of scientific uses (PDL is not a substitute for Numpy/SciPy/Pandas).

Don't get me wrong. I own at least 12 Perl books and have reviewed some of the latest Perl 6 (Raku) books and really really like Raku. There's not much wrong with Perl in my book (certainly not what most people claim). I've reached for it a few times at work and found it to be fairly pleasant to use. My biggest complaints about building complex data structures in Perl (yes it's much easier than C, but still complicated if you're coming from Python) and having to remember when something is "$" that in my mind should be "@" have all been fixed by Raku.

I just wanted to point out that your views about it being much faster across the board don't seem to be correct.

Any data to back up this claim? My experience in dealing with large numbers of files, huge data sets, is that Perl generally roars, while python lags. I converted a python code to Perl associated with some recent work, and got a good ~2.5x overall improvement. It varied a bit as a function of the input, but generally Perl is superior in performance. Some areas Python is better, but not many.

> I don't think your performance numbers are very accurate.

Food for thought.

https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

More food for thought: https://bitbucket.org/ewanhiggs/csv-game/src/master/README.m...

So Python is faster than Perl on 6/10 benchmarks? This seems to support my argument that they are generally fairly comparable and definitely not that Perl is waaay faster.

No, the numbers are pretty accurate. In most of the cases I've run into, Perl is faster than Python for its operations. Some trivial examples[1] showed up on HN a few weeks ago, about how this person "optimized" python, by eventually replacing python with C.

As for the nested complex data structures, Perl's been doing that forever (e.g. since 5.x started). And its trivial to use.

That's one of the nicer aspects of Perl. Things you think should work, often, just do. And work the way you want them to. It's not perfect. It is very, very good though.

[1] https://scalability.org/2020/05/on-optimizing-scripting-lang...

Don't forget about PyPy, which may well be faster than Perl (I haven't compared them) or Cython if you're stuck with CPython.

That's a good point.

I tried googling around and couldn't find any speed comparisons between the languages, just talk between the two like eg https://news.ycombinator.com/item?id=8626131

So, I guess I'll address this directly:

The tasks Perl is designed to do are different than the kinds of tasks PyPy is designed to do, so it is a bit of apples an oranges comparison.

PyPy starts to get its speed benefit when a task takes 1) longer than 3 seconds to run and 2) Is directly written in Python, not using libraries written in C.

Perl is designed to parse large files, do string parsing, and one off scripts. It being fast is nice, but it's not designed to be running large math heavy processing scripts to begin with. If a Perl script takes longer than a second to run, not including IO limitations and network limitations, it may not be the ideal tool for the job.

So while PyPy gets comparable to Perl speeds, Perl being fast is almost moot, because it's not designed to be used that way to begin with.

Reply on news.ycombinator.com