Looks really interesting. PHP may not be the "hip" language to use (although it's getting better), but it's so popular and accessible that disregarding it as a legitimate option for new projects is myopic.
In some ways they've introduced some powerful features bringing it on par with other languages, but the reality is that performance varies wildly in the real world because opcaches are finicky dependent on the data/request type/frequency and automated GC/circular-reference optimization is far from perfect.
Between HHVM (to effectively make memory growth linear vs logarithmic when approaching higher request concurrency) and sensible opcache tweaks you can make that runtime a bit less stochastic at scale, however if you don't know what you're doing PHP makes it so easy to go in the complete opposite direction without any indications you're running toward a cliff...
What does it mean for a language to have memory leaks? Are you talking about a problem specific to the PHP runtime or common PHP idioms that lead to leaky code? A specific example would be helpful as I've never known PHP to be particularly vulnerable to memory-leak bugs (at least not any more than any of the other popular imperative scripting languages).
Also, what's the substance behind the "etc" in your comment? I'm not really sure what naturally follows in the sequence of "memory-leaks and so on...."
Your comment seems to simply cast doubt without much substance.
I'm not embroiled in the PHP-world, so I don't know how many alternate PHP runtimes exist, but I presume that the default PHP runtime itself was being referenced. Seeing as for a long time there was a single runtime paired with the language (I'm 99% sure that PHP4 didn't have alternate runtimes), I think that equating bugs in the runtime with bugs in the language is something that happens even if technically they are separate things.
> Your comment seems to simply cast doubt without much substance.
And your comment seems needlessly defensive. I made a comment about general sentiment that I've gotten from people that have actual working knowledge of large-scale PHP deployments. I'm sorry that in the middle of the conversation I didn't start immediately interrogating these people asking why they haven't filed bug reports if there are memory leaks.
Also, is it devoid of substance if I'm basically stating that I've talked to people "in the trenches" (and presumably know what they are talking about) who don't have the opinion that PHP is some beautiful, but misunderstood language that gets a bad rap?
> Also, what's the substance behind the "etc" in your comment? I'm not really sure what naturally follows in the sequence of "memory-leaks and so on...."
Maybe I should have phrased that as "(e.g. memory leaks)" to show that memory leaks weren't the only complaint, but my memory fails me on what the other complaints were.
Yes, it's true that there's a ton of really bad code out there. But that's a measure of it's popularity.
There's also a ton of bad C and Java out there.
And yet, no one dismisses those languages because of that.
(But don't worry, dear reader -- you're probably the exception!)
PHP gets singled out for a bit of extra abuse (some perhaps justifiably, some not so much), but it's far from the only language whose popular conception is not so much based on thoughtful evaluation.
So are you talking about the default PHP 4 runtime or the default PHP 5.3 runtime or the default PHP 5.4 runtime etc? These versions are very different beasts, so levying a vague "memory leak" criticism at PHP without substantiating it via code or even a version makes the criticism pretty meaningless.
> I think that equating bugs in the runtime with bugs in the language is something that happens even if technically they are separate things.
I asked for clarification because I was unsure what you were referring to. For example, about a month ago I was tasked with debugging a memory leak in a legacy Ruby application where I eventually discovered that the developer had used a proc in such a way that it prevented the GC from freeing variables that had been referenced in the closure. It was an issue somewhat similar to the examples described in this blog post:
http://victorarias.com.br/2013/08/13/leaky-ruby.html
I wouldn't attribute this memory leak to the Ruby langauge but more to a Ruby idiom that developers should take care to avoid. That's the type of clarification I'm looking for, but it appears that you don't really have anything to offer the discussion in this regard.
> And your comment seems needlessly defensive. I made a comment about general sentiment that I've gotten from people that have actual working knowledge of large-scale PHP deployments. I'm sorry that in the middle of the conversation I didn't start immediately interrogating these people asking why they haven't filed bug reports if there are memory leaks.
I have no emotions tied up in this discussion, I'm trying to make a point that your criticism is just anecdotal hearsay without any way for someone to verify if what you're saying is actually true. I didn't suggest that you ask your colleagues to file bug reports, I was simply pressing you for more details since your vague anecdote runs counter to my own experiences with PHP as I relate it to the rest of the imperative scripting landscape.
> Also, is it devoid of substance if I'm basically stating that I've talked to people "in the trenches" (and presumably know what they are talking about) who don't have the opinion that PHP is some beautiful, but misunderstood language that gets a bad rap
Yes, that is pretty much the definition of "devoid of substance". "In the trenches" is a subjective description that doesn't mean anything. You presume they know what they're talking about, but maybe they actually don't, but we'll never know either way because all you've done is recount nebulous secondhand generalizations. You're not even asking me to take your word for it, you're asking me to take your word for their word. Can't you see why that's a little dubious?
It was a big deal when transitioning to the FPM runtime. That's why persistent PHP runtimes like FPM still have the option to completely reinitialize itself after a set number of pageviews. It's a bit like fixing leaks by rebooting, but it's fast enough not to be noticeable.
It's also present in Apache, and thus mod_rails and mod_wsgi, by default[1], is an option in Gunicorn[2], and a gem is also available to provide the same in Unicorn.[3]
1 http://httpd.apache.org/docs/2.2/mod/mpm_common.html#maxrequ...
2 http://gunicorn-docs.readthedocs.org/en/latest/settings.html...
I've used PHP a few times in the past for some long-running, complex cron jobs, and I never had memory leaks. I've also never noticed it for short-running HTTP responses, and some of those have been really awful and messy because I'm more of a PHP doctor than someone who starts projects with it.
Anyway, everyone is going to benefit from HHVM. If Facebook uses it in production, you can be sure that it won't have any common memory leaks for long.
IIRC arrays are always passed by reference and only duplicated if you modify the data inside the called function (if not passed with '&' of course) ?
Duplicating an array in memory every time it is passed to or from a function would be awful.
This might be because I originally learned PHP about 16 years ago, but I could swear this used to be the case. There's definitely a reason PHP has the reputation it has, although many of those glaring issues have since been fixed.
> only duplicated if you modify the data inside the called function
I did not know that. Because I've observed the duplication behavior, I assumed that it was never a reference. I guess that's a "feature" when you don't really have any fine-grained, explicit control over pointers or references.