18

A few do, but not any of the popular ones as far as I know. Is there something bad about nesting comments?

I plan to have block comments nest in the (small) language I'm working on, but I would like to know if this is a bad idea.

  • re a few answers: ohh, that makes sense =) I'm totally doing nested block comments then; although I do have a separate lexing stage, it's not the limiting sort SK-logic described. – Vuntic Jun 2 '11 at 11:05
  • @Vuntic: If you have a separate lexing stage that uses stuff more complicated than regular expressions, you may have performance issues. REs are fast and easy to use by implementing DFAs. – David Thornley Jun 2 '11 at 16:18
  • It catches more errors earlier to not allow nesting – user1249 Jun 2 '11 at 16:21
  • 4
    @David: ...not at all. It's actually really fast. – amara Jun 2 '11 at 20:28
  • I would suggest that if you want to allow nested comments, you allow start-comment tags to be marked with a token, and require that if a start-comment tag is thus marked, its end-comment tag must be marked identically. That would allow unbalanced start/end tags to be quickly identified, and avoid the possibility of bugs caused by undetected unbalanced tags. – supercat May 16 '14 at 20:40
6

One thing nobody's mentioned yet, so I'll mention it: The desire to nest comments often indicates that the programmer is Doing It Wrong.

First, let's agree that the only time "nesting" or "not nesting" is visible to the programmer is when the programmer writes something structurally like this:

do_something();
/* comment /* nested comment */ more comment */
do_something_else();

Now, when does such a thing come up in practice? Certainly the programmer isn't going to be writing nested comments that literally look like the above snippet! No, in practice when we nest comments (or wish we could nest them), it's because we want to write something like this:

do_something();  /* do a thing */
/* [ajo] 2017-12-03 this turned out to be unnecessary
do_something_else(); /* do another thing */
*/

And this is BAD. This is not a pattern we (as language designers) want to encourage! The correct way to write the above snippet is:

do_something();  /* do a thing */

That "wrong" code, that false start or whatever it was, doesn't belong in the codebase. It belongs, at best, in the source-control history. Ideally, you'd never even write the wrong code to begin with, right? And if the wrong code was serving a purpose there, by warning maintainers not to reinstate it for some reason, well, that's probably a job for a well-written and intentional code comment. Trying to express "don't do X" by just leaving in some old code that does X, but commented out, is not the most readable or effective way to keep people from doing X.

This all boils down to a simple rule of thumb that you may have heard before: Don't comment out code. (Searching for this phrase will turn up a lot of opinions in agreement.)

Before you ask: yes, languages such as C, C#, and C++ already give the programmer another tool to "comment out" large blocks of code: #if 0. But this is just a particular application of the C preprocessor, which is a large and useful tool in its own right. It would actually be extremely difficult and special-casey for a language to support conditional compilation with #if and yet not support #if 0.


So, we've established that nested comments are relevant only when the programmer is commenting out code; and we've established (via consensus of a lot of experienced programmers) that commenting out code is a Bad Thing.

To complete the syllogism, we must accept that language designers have an interest in promoting Good Things and discouraging Bad Things (assuming that all else is equal).

In the case of nested comments, all else is equal — you can safely ignore the low-voted answers that claim that parsing nested /* would somehow be "difficult" for the parser. (Nested /* are no harder than nested (, which just about every parser in the world already needs to handle.)

So, all else being equal, should a language designer make it easy to nest comments (i.e., to comment out code), or difficult? Recall that commenting out code is a Bad Thing.

Q.E.D.


Footnote. Notice that if you don't allow nested comments, then

hello /* foo*/bar.txt */ world

is a misleading "comment" — it's equivalent to

hello bar.txt */ world

(which is likely a syntax error). But if you do allow nested comments, then

hello /* foo/*.txt */ world

is a misleading "comment" — it's equivalent to

hello

but leaves the comment open all the way to the end of the file (which again is almost certainly a syntax error). So neither way is particularly less prone to unintentional syntax errors. The only difference is in how they handle the intentional antipattern of commented-out code.

  • 1
    I have different opinion based on simply fact -- I didn't see everything (and neither did you). So while those golden rules like "Don't comment out code" look nice, life has its own paths. In this particular case, I do it very often as as switch, when I test some new feature and have to incrementally introduce some code, so I comment out the code, then less, less, less, and finally I have working piece and I can remove all the comments (over code). My perfect language will of course support nested comments :-). – greenoldman Dec 4 '17 at 16:07
  • @greenoldman: Most languages don't have nestable comments, but they'll have some actual feature for "remove a block of code" which is less-used than the "leave a comment" feature. C's #if DEAD is the canonical and best-designed example. In many languages you can just wrap the dead code in the equivalent of if (DEAD). And in many IDEs, you can actually remove the dead code and rely on Ctrl+Z and/or version control to get it back if you want it. Leaving a comment, docstring, whatever, whose text is a bunch of dead code, is still the worst option for readability. – Quuxplusone Dec 4 '17 at 19:10
11

Because most of the implementations are using separate lexing and parsing stages, and for lexing they're using plain old regular expressions. Comments are treated as whitespaces - i.e., ignored tokens, and thus should be resolved entirely in a lexing pass. The only advantage of this approach is parsing speed. Numerous disadvantages include severe limitations on syntax (e.g., a need to maintain a fixed, context-independent set of keywords).

  • 3
    I would disagree on 'most' nowadays. Certainly that's the traditional way, but I know that for C, EDG combines the preprocessor, lexing and parsing, and I suspect that both GCC and Microsoft do too. The benefit is that it allows you to implement them separately if you need to. – Andrew Aylett Jun 2 '11 at 9:56
  • Clang is doing the same, too. But that's still only a tiny proportion of the existing popular languages compilers. – SK-logic Jun 2 '11 at 10:00
  • @Neil Butterworth, take a look at mcs, javac, gcc (yep, it back-patches a lexer, but still it is a dedicated lexing pass), clang (same as gcc), dmd, fpc, and many, many more. – SK-logic Jun 4 '11 at 23:09
  • No one is using regular expressions in their lexing for any non-trivial compiler. – Nuoji Jan 10 at 11:47
  • @Nuoji - for the non-trivial - sure. But those who rely on flex and similar tools do. – SK-logic Jan 11 at 15:36
7

It's perfectly possible to make a lexer that can handle nested comments. When it's eating whitespace, when it sees /* it can increment a depth counter, and decrement it when it sees */, and stop when the depth is zero. That said, I've done many parsers, and never found a good reason for comments to nest.

If comments can nest, then a downside is it's easy to get their ends unbalanced, and unless you have a fancy editor, it can invisibly hide code you assume is there.

An upside of comments that don't nest is something like this:

/*
some code
more code
blah blah blah
/**/

where you can easily comment the code in or out by removing or adding the first line - a 1-line edit. Of course, if that code itself contains a comment, this would break, unless you also allow C++-style // comments in there. So that's what I tend to do.

  • 1
    // comments are also C99-style. – JAB Jun 2 '11 at 20:58
  • Alternatively, a language could specify a start-of-comment is /*$token, where identifier is any alphanumeric token, and end-of-comment is token$*/. It would be relatively simple for tokenizer to include code to verify that every end-comment mark contains the proper token for its matching start-comment block. – supercat May 16 '14 at 20:36
5

Since nobody else mentioned it, I'll list a few languages that do support nested comments: Rexx, Modula-2, Modula-3, Oberon. Despite all the complaints here about difficulty and speed issues, none of those seem to have any huge problems.

  • 4
    To which I add: Haskell, Frege – Ingo Aug 24 '11 at 9:31
  • Supported by Scala too. – Matt R Aug 13 '13 at 15:23
4

A good point of nesting block comments is that you can comment out large portions of code easily (well, almost, unless you have the block comment end sequence in a string constant).

An alternative method is to prepend a bunch of line with the line comment start sequence if you have an editor that supports it.

Haskell has nested block comments, but most people dont seem to notice or to complain about it. I guess this is because people that do not expect nested comments tend to avoid them as this would be a lexical error in other languages.

3

Supporting nested block comments complicates the parser, which is both more work and it could increase the compile time. I guess it is not a very needed feature for a language, so it is better to use the time and effort on other improvements and optimizations.

In my opinion simplicity is always a good thing in designing anything. Keep in mind that it is easier to add a feature than to remove it. Once you allow nested comments and there are programs out there using it, you won't be able to take them out without breaking compatibility.

  • 1
    +1 for "easier to add a feature than to remove it". – R.. Jun 8 '11 at 0:49
  • 3
    once you disallow nested comments you can't allow them as well because it'll break such comments: /*/**/ – RiaD Jun 19 '14 at 20:23
2

One probable reason is that nested comments must be handled by the parser, since the flavor of regular expressions commonly used in lexers don't support recursion. The simple ones can be eliminated as whitespace by the lexer, so they're simpler to implement in that way.

  • 3
    It's not the "flavor". The word "regular" in regular expression inherently excludes recursion. – R.. Jun 8 '11 at 0:48
  • 3
    @R: In mathematics, sure. But in programming, we have things that we call regexes that do support recursion. – amara Jun 8 '11 at 19:33
  • The question is: Is this even an issue? Most languages already have to deal with nesting parentheses. To name some: Lisp, C, Java, Python, Ruby, Perl. – Thomas Eding Apr 23 '14 at 20:34
  • Nested parentheses are fine, because the things inside the parentheses are the same as the stuff outside: normal tokens. In comments, you don't have tokens, you just have text. You need to be able to match the start and end comment tokens so that you know whether 'int' is a type or just a word in a comment. (Especially if you eliminate comments in the lexer.) – Alan Shutko Jun 19 '14 at 19:12
  • 2
    @ThePopMachine: I'm sure of what I stated, that regular has a defined formal meaning, not the meaning you're using, and that the "regular" in "regular expression" was chosen for this meaning. Being non-recursive is one result of its definition. – R.. Jun 19 '14 at 20:57
-1

Who knows? I would guess because supporting nested comments is more work - you would have to maintain a stack of some sort, and because it complicates the language grammar.

-1

Nested comments mean extra work for the parser. Usually when you see the start of a comment you ignore everything until the end comment marker. In order to support nested comments you have to parse the text in the comments as well. The biggest issue, though, is that a programmer has to be careful to close all nested comments correctly or it will lead to compilation errors. Correctly implementing a compiler is something that can be done but keeping track of nested comments as a programmer is quite error-prone and irritating.

  • 3
    -1: not true. Sane parsers don't work like that. – amara Jun 8 '11 at 19:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.