This is one of the features that Ruby cribbed directly from Perl. The Ruby documentation seems really bad, in particular “interpolation mode” is grievously misleading.
Perl’s documentation is far more clear about the consequences:
o Compile pattern only once.
[…]
PATTERN may contain variables, which will be
interpolated every time the pattern search is
evaluated, except for when the delimiter is a
single quote. […] Perl will not recompile the
pattern unless an interpolated variable that
it contains changes. You can force Perl to skip
the test and never recompile by adding a /o
(which stands for "once") after the trailing
delimiter. Once upon a time, Perl would recompile
regular expressions unnecessarily, and this
modifier was useful to tell it not to do so,
in the interests of speed. But now, the only
reasons to use /o are one of:
[reasons]
The bottom line is that using /o is almost
never a good idea.
In the days before Perl automatically memoized the compilation of regexes with interpolation, even back in the 1990s, it said,
However, mentioning /o constitutes a promise
that you won't change the variables in the
pattern. If you change them, Perl won't even
notice.
PATTERN may contain references to scalar
variables, which will be interpolated
(and the pattern recompiled) every time the
pattern search is evaluated. […] If you want
such a pattern to be compiled only once, add
an “o” after the trailing delimiter. This
avoids expensive run-time recompilations, and
is useful when the value you are interpolating
won't change over the life of the script.
This is a footgun. A language should strive not to add footguns. Every footgun you provide, somebody is going to blow their foot off with it, so that's a high price. If your language is popular it might be a lot of somebodies.
The opposite behaviour (we have a constant regular expression, we re-use it often but the tooling doesn't realise and so it's created each time we mention it) is not a footgun, it results in poor performance, and so you might want (especially in some managed languages) to just magically optimise this case, but if not you won't cause mysterious bugs. An expert, asked "Why is this slow?" can just fix it - you have to supply basic tools for that, but this flag is not a sensible tool.
In the spirit of "what's old is new again," PowerShell also has the same idea, and is done per Function with "begin", "process", "end", and "clean" stanzas that allow setup, teardown, for-each-item, and "finally" behavior: https://learn.microsoft.com/en-us/powershell/module/microsof...
Oh, that’s an interesting take. I’ve long been looking for newer developments on Awk’s clause structure, and this seems like an interesting take (though I’m unclear on whether I can have multiple begin/end clauses, which are the best thing about Awk’s version). It also finally connects this idea to something else in my mind—specifically advice[1] and CLOS’s :before/:after/:around methods[2]. (I guess Go’s defer also counts?)
function Fred {
begin {
echo "hello from begin1"
}
begin {
echo "hello from begin2"
}
process {
echo "does the magic"
}
}
$bob = @("alpha" "beta")
$bob | Fred
Then
$ pwsh fred.ps1
ParserError: /Users/mdaniel/fred.ps1:5
Line |
5 | begin {
| ~~~~~~~
| Script command clause 'begin' has already been defined.
> Modifier o means that the first time a literal regexp with interpolations is encountered, the generated Regexp object is saved and used for all future evaluations of that literal regexp.
That is crystal clear to me. It means that on the next execution, the new values of the interpolation will be ignored; the regexp is now "baked" with the first ones.
Like this in C++:
void fun(int arg)
{
static int once = arg;
}
if we call this as f(42) the first time, once gets initialized to 42. If we then call it f(73), once stays 42.
There is a function in POSIX for once-only initializations: pthread_once. C++ compilers for multithreaded environments emit thread-safe code to do something similar to pthread_once to ensure that even if there are several concurrent first invocations of the function, the initialization happens once.
This isn't the same problem, because this is about whether the regex is instantiated each time the code around the regex is executed, or only the first time and cached for subsequent executions. The same could in theory happen with closures, but I haven't ever seen programming-language semantics where, for example, a function containing the definition of a closure that depends on an argument of that outer function, would use the argument value of the first invocation of the function for all subsequent invocations of the function.
For example, when you have
fn f x = (y -> x + y)
then a sequence of invocations of f
f 1 3
f 2 6
will yield 4 and 8 respectively, but never will the second invocation yield 7 due to reusing the value of x from the first invocation. However, that is precisely what happens in the article's regex example, because the equivalent is for the closure value (y -> x + y) to be cached between invocations, so that the x retains the value of the first invocation of f — regardless of whether x is a reference by name or by value.
The parallel is apt, but regex /o is more like a closure that captures by value at declaration time rather than an ambiguity between capture strategies.
The context is that this is a feature cribbed straight from perl, where where it's passed down from perl 4/pre-5.6, where compiled regexen weren't first-class values. Pretty much every use of it this century is a mistake.
Yeah but it's kind of tired when it's being used every time someone makes a mistake with regex. I've used them extensively in my career and never once regretted it.
The problem with regexps is that “Sometimes a smart person, who has done the work, and knows how to leverage regular expressions correctly, decides they are appropriate for solving a problem where there is shared maintenance. Now, you have people who haven’t put in the work, and have been told repeatedly through ‘witty quips’ to not bother.”
> I didn’t recognize /o. It didn’t seem critically important to lookup yet.
> With nothing else to investigate, I finally looked up the docs for what the /o regex modifier does.
I'll probably never understand this mode of thinkning. But then again, Ruby programmers are, after all, people who chose to write Ruby.
> /o is referred to as “Interpolation mode”, which sounded pretty harmless.
Really? Those words sound quite alarming to me, due to personal reminiscences of eval.
Also, this whole "/o" feaure seems insane. If I have an interpolation in my regex, obviously I have to re-interpolate it every time a new value is submitted, or I'd hit this very bug. And if the value is expected to the same every time, then I can just compile it once and save the result myself, right? In which case, I probably could even do without interpolation in the first place.
It's a feature dating from the 1990s, when Perl (and I guess Ruby?) didn't have a way for the user to store a compiled regex, and this was a useful shortcut for a very specific optimization, which Ruby documented badly. Perl (and I guess Ruby?) later evolved in a way that made /o unnecessary, but the (now mis)feature remained.
“Compilation”, I think, is exactly right. This feature is less about interpolation than it is about compilation of a single regexp to be used many times. It’s just shrouded in confusing documentation that should say: “/o tells ruby to rewrite this code such that it refers to a new statically allocated regexp object.” And when you write it that way, you see how insane it is for a function call to be hoisted automatically like this, without an explicit, obvious, syntactic annotation.
This is one of the features that Ruby cribbed directly from Perl. The Ruby documentation seems really bad, in particular “interpolation mode” is grievously misleading.
Perl’s documentation is far more clear about the consequences:
(https://perldoc.perl.org/perlop#Regexp-Quote-Like-Operators)
In the days before Perl automatically memoized the compilation of regexes with interpolation, even back in the 1990s, it said, Perl 4’s documentation is briefer. It says,(https://github.com/Perl/perl5/blob/perl-4.0.00/perl.man#L272...)
https://perldoc.perl.org/perlre
This is a footgun. A language should strive not to add footguns. Every footgun you provide, somebody is going to blow their foot off with it, so that's a high price. If your language is popular it might be a lot of somebodies.
The opposite behaviour (we have a constant regular expression, we re-use it often but the tooling doesn't realise and so it's created each time we mention it) is not a footgun, it results in poor performance, and so you might want (especially in some managed languages) to just magically optimise this case, but if not you won't cause mysterious bugs. An expert, asked "Why is this slow?" can just fix it - you have to supply basic tools for that, but this flag is not a sensible tool.
Unsurprisingly, `END {}` is also inherited from perl, tho I think it originally comes from awk.
Similarly unsurprisingly, with its BEGIN friend https://docs.ruby-lang.org/en/3.3/syntax/miscellaneous_rdoc....
In the spirit of "what's old is new again," PowerShell also has the same idea, and is done per Function with "begin", "process", "end", and "clean" stanzas that allow setup, teardown, for-each-item, and "finally" behavior: https://learn.microsoft.com/en-us/powershell/module/microsof...
Oh, that’s an interesting take. I’ve long been looking for newer developments on Awk’s clause structure, and this seems like an interesting take (though I’m unclear on whether I can have multiple begin/end clauses, which are the best thing about Awk’s version). It also finally connects this idea to something else in my mind—specifically advice[1] and CLOS’s :before/:after/:around methods[2]. (I guess Go’s defer also counts?)
[1] https://en.wikipedia.org/wiki/Advice_(programming)
[2] https://gigamonkeys.com/book/object-reorientation-generic-fu...
It seems not:
Given:
ThenAs an old Perl programmer, I knew immediately what the /o would do. ;-)
I've always loved the recent[1] summary from `perlre`:
>o - pretend to optimize your code, but actually introduce bugs
1: I still think of it as a relatively new change, but it's from 2013: <https://github.com/Perl/perl5/commit/7cf040c1f649790a4040aec...>
It's older than that. The article links to this conversation about it in 2003: https://www.perlmonks.org/?node_id=256053
It looks like an emoji for someone getting bashed in the head with a long stick. So that makes sense?
> Modifier o means that the first time a literal regexp with interpolations is encountered, the generated Regexp object is saved and used for all future evaluations of that literal regexp.
That is crystal clear to me. It means that on the next execution, the new values of the interpolation will be ignored; the regexp is now "baked" with the first ones.
Like this in C++:
if we call this as f(42) the first time, once gets initialized to 42. If we then call it f(73), once stays 42.There is a function in POSIX for once-only initializations: pthread_once. C++ compilers for multithreaded environments emit thread-safe code to do something similar to pthread_once to ensure that even if there are several concurrent first invocations of the function, the initialization happens once.
Seems par for the course for Ruby.
Love these sorts of deep dives, thanks!
This is the same problem people have with closures, where it's unclear to the user whether the argument is captured by name or by value.
This isn't the same problem, because this is about whether the regex is instantiated each time the code around the regex is executed, or only the first time and cached for subsequent executions. The same could in theory happen with closures, but I haven't ever seen programming-language semantics where, for example, a function containing the definition of a closure that depends on an argument of that outer function, would use the argument value of the first invocation of the function for all subsequent invocations of the function.
For example, when you have
then a sequence of invocations of f will yield 4 and 8 respectively, but never will the second invocation yield 7 due to reusing the value of x from the first invocation. However, that is precisely what happens in the article's regex example, because the equivalent is for the closure value (y -> x + y) to be cached between invocations, so that the x retains the value of the first invocation of f — regardless of whether x is a reference by name or by value.The parallel is apt, but regex /o is more like a closure that captures by value at declaration time rather than an ambiguity between capture strategies.
It's kind of a cool feature. I like it.
Is it? I can't think of a non-contrived case where this would actually be useful.
And in any case where it would be useful, it seems like a better way to optimize would just be to refactor the regex out into a constant.
The context is that this is a feature cribbed straight from perl, where where it's passed down from perl 4/pre-5.6, where compiled regexen weren't first-class values. Pretty much every use of it this century is a mistake.
Actually, I have a way this would work well. If you’re interpolating a value that comes from configuration and wouldn’t change.
Example: /admin@#{Rails.config.x.domain}/io
But you’re right that a constant would be a lot more clear. “o” is a footgun.
Im sorry but the classics never go out of style:
"Some people, when confronted with a problem, think 'I know, I'll use regular expressions.' Now they have two problems."
Yeah but it's kind of tired when it's being used every time someone makes a mistake with regex. I've used them extensively in my career and never once regretted it.
The problem with regexps is that “Sometimes a smart person, who has done the work, and knows how to leverage regular expressions correctly, decides they are appropriate for solving a problem where there is shared maintenance. Now, you have people who haven’t put in the work, and have been told repeatedly through ‘witty quips’ to not bother.”
The second problem being how to deal with all the extra time they just freed up?
> I didn’t recognize /o. It didn’t seem critically important to lookup yet.
> With nothing else to investigate, I finally looked up the docs for what the /o regex modifier does.
I'll probably never understand this mode of thinkning. But then again, Ruby programmers are, after all, people who chose to write Ruby.
> /o is referred to as “Interpolation mode”, which sounded pretty harmless.
Really? Those words sound quite alarming to me, due to personal reminiscences of eval.
Also, this whole "/o" feaure seems insane. If I have an interpolation in my regex, obviously I have to re-interpolate it every time a new value is submitted, or I'd hit this very bug. And if the value is expected to the same every time, then I can just compile it once and save the result myself, right? In which case, I probably could even do without interpolation in the first place.
It's a feature dating from the 1990s, when Perl (and I guess Ruby?) didn't have a way for the user to store a compiled regex, and this was a useful shortcut for a very specific optimization, which Ruby documented badly. Perl (and I guess Ruby?) later evolved in a way that made /o unnecessary, but the (now mis)feature remained.
“Compilation”, I think, is exactly right. This feature is less about interpolation than it is about compilation of a single regexp to be used many times. It’s just shrouded in confusing documentation that should say: “/o tells ruby to rewrite this code such that it refers to a new statically allocated regexp object.” And when you write it that way, you see how insane it is for a function call to be hoisted automatically like this, without an explicit, obvious, syntactic annotation.
The implications of "statically allocated" are less clear than if you'd just write "compiled only once".