Comments should be sentences | Hacker News Reader

Comments should be sentences

nedbatchelder.com

42 pointshenrik_w49 comments

Never write comments that re-state the implementation.

    # Try to read the configuration file.

If this wasn't already obvious without the comment, then the code is a mess. Tidy the code up instead of painting over the problem with comments.

That's a rule to be applied with a bit of nuance. Some things, especially in languages like Perl whose syntax already reads like a half-deliberate attempt to force an obscurantist style on the programmer, can't be made really obvious without doubling or tripling the length of the code in which they're implemented, something which imposes its own cost in readability and maintainability. In cases like that, it can be useful to add a few comment lines which briefly restate what the code implements, as both an aide-memoire for the original developer, and a minimal but immediately accessible answer to the question of "what the hell is this doing?" on the part of those who follow.

I agree wholeheartedly. The syntax doesn't even have to be extremely obscure, it can be highly useful in lower-level languages such as C as well: Often, I just want to get a quick idea of what the next big block of code is going to do without having to read the entire block of code.

In that case it's often (yea I know, not always) better to put that block in a separate function, like readConfigurationFile() instead of "// This block reads the configuration file.". I almost always treat the act of writing comments as a signal that my code is a mess...although they're sort of a necessary evil when you're down to crunch time.

I agree, that's an excellent reason to create a function/method. Reason 7, self-documenting code: http://henrikwarne.com/2013/08/31/7-ways-more-methods-can-im... (but of course it doesn't mean that you never need comments)

I think it's said that you should document the "why" (as opposed to "what"), which is relevant when you're using an unconventional approach or for some other reason your code may not be understood immediately, or when business logic is involved.

A fair point, I should probably have mentioned that. Thanks for pointing that out.

I would think that a comment re-summarizing the function in natural language would also "double or triple the length of the code." Wouldn't it be better, then, to just let the code itself expand, rather than expecting maintenance programmers to both read your comments, and then use them to parse an otherwise-inscrutable construction?

> I would think that a comment re-summarizing the function in natural language would also "double or triple the length of the code."

It needn't; the example I had in mind when I wrote my earlier comment was some Perl code I wrote on Friday to automate the task of taking one database, full of relations dependent on auto-increment column values, and merge its contents into another database of identical schema but conflicting auto-increment values, while maintaining the relationships between rows. This worked out to ca. 60 lines of Perl, and unusually dense and hard-to-read lines even for that language; to this, I added about a half-dozen terse but informative comment lines to serve as guideposts for my primarily Python-hacking colleague and co-conspirator on this project.

A 10:1 code-to-comments ratio is unusually high, granted, and perhaps I could've instead written the code to use a lot more temporary variables and otherwise ballooned it out to twice or three times its length, but I don't see how that would've aided readability; on the one hand, there'd be a lot more state to keep track of, and on the other, it'd no longer fit on a single screen, and would thus require the reader to scroll hither and yon while trying to make sense of it.

Some things are inherently not simple, and I'd argue that complex manipulations of complex data structures fall well within that category -- in such cases, there's only so far you can simplify the algorithm before it ceases to work as you intend. When you write such code, you more or less have to rely on the native competence of whoever else has to work with it, because there's only so much hand-holding you can accomplish either in comments or in code.

Fortunately, in this case I can safely so rely; my colleague is at least as capable as I am, so the comments I added are the same sort I'd appreciate having when attempting to digest a complex algorithm in a language not wholly familiar to me -- for example, if he'd written this code in Python, rather than me writing it in Perl.

(Of course, the real problem is that, while both languages make it possible to perform the sort of manipulation necessary here, neither language is really well suited to it; if my colleague knew Lisp as well as I do, I could've written the bloody thing in fifteen elegant and highly readable lines, and we both could go on with our lives. But I've more or less given up hope of introducing Lisp at my place of work; even the most seasoned programmers among my colleagues, when exposed to parentheses, turn pale green and make excuses to be somewhere else in a hurry.)

I think another exception (in addition to the already mentioned help with an otherwise difficult syntax) is when comments are used to visually represent logical segments of the code. Some style guides suggest to write these in all caps. Eg:

  # CONFIG
  ...
  # LOCALS
  ...
  # EDGE CASES
  ...
  # REFORMAT INPUT
  ...
  # MAIN ALGORITHM
  ...

Etc.

Rather than an exception, I think this is an example of the precise scenario in which it's most inappropriate to use comments.

    > visually represent logical segments of the code

This is exactly the use case in which function definitions excel. Use functions for this, not comments.

Yeah. A past HN discussion caused me to realize that comments are often used for two independent purposes: structure and commentary. Conflating the two has caused many conversations with the participants talking past each other.

http://akkartik.name/post/2012-11-24-18-10-36-soc

Maybe this is because I come from an HTML + CSS background, but if I'm adding code to a file that may need to be moved around within that file or transplanted to another file I often use comments to mark the start and finish of Sachs ection or 'block' of code.

It has little to do with how clean the code is, I can't imagine not having my way finding markers around!

Such commentary goes at the top of a function or class. You get more nuanced as you delve deeper into the textual abyss.

Agreed. I often try to make the point that when commenting people in my group should focus more on commenting on WHY and HOW rather than WHAT that is normally easily read from the code.

I only started doing this recently. It occurred to me that I had some conception in my subconscious that natural language punctuation wasn't required in formal language code files. I think it may have simply been a cognitive spill-over from the fact that variable names start with lower-case characters.

Looking back, it's totally ridiculous that I would, in a professional setting, put my name next to unpunctuated writing. Aside from being ugly and hard to read, it's straight-up unprofessional.

As of 2 months ago Ned doesn't follow his own advice:

https://github.com/nedbat/byterun/blob/master/byterun/execfi...

What're you referring to? They all seem to be complete sentences with capitals and full-stops.

Sorry, I didn't think it was hard to find these:

     13     open_source = tokenize.open     # pylint: disable=E1101
     27	# from coverage.py:
     30	    # In Py 2.x, the builtins were in __builtin__
     33	    # In Py 3.x, they're in builtins
     38	    """The same as s.rsplit(sep, 1), but works in 2.3"""
    101	    # Create a module to serve as __main__
    140	        # Restore the old __main__
    143	        # Restore the old argv and path

A reminder of what the blog post tells us:

     Start comments with a capital, have a subject and a
     verb, and end them with a period.

http://en.wikipedia.org/wiki/Tu_quoque

http://en.wikipedia.org/wiki/Straw_man

Cue the self-documenting, we-don't-need-no-stinking-comments rabble rousers in 3, 2, 1...

AND that's one reason why we're not an actual profession, people.

We aren't?

See http://michaelochurch.wordpress.com/2012/11/18/programmers-d...

That's adding some rather subjective baggage to the term "profession".

I think the only thing really stopping software development from becoming a formal profession is a standardized code of ethics (of which there are several proposed) and a professional body to govern it (and let's face it, professional bodies can exert power to satisfy immoral agendas, much like unions).

Then:

Professionals are allowed and expected to dedicate about half their working time to career development and continuing education. Many software engineers will get fired if they get caught in that.

This is pretty much false.

Honestly, given all of this, I don't even think software development becoming a profession is desirable at all.

So, we aren't then, and we don't want to be. Yep, that's pretty much what I'm saying.

My paycheck begs to differ!

But really, comments are one way to mislead in code. If a comment is made that incorrectly states the intent of code, it's undeniable that this is harmful to the readability of the code.

What's pay got to do with it? See

http://michaelochurch.wordpress.com/2012/11/18/programmers-d...

Comments may mislead yes, but that's a problem with any sufficiently expressive language, and code is one such language: code can mislead. The functionality required of the code may be only half-implemented, then that code is misleading. That code may work for certain easy cases, but it may be harmful to your understanding of what functionality is actually required.

It's easy to see why we prioritize code over comments, why we prioritize the mechanisms over the specifications. It's because code can have some manner of automatic verification, and we're too harried to look into the comments, to do code review. Deadlines abound. We can't stop work and say, no, we can't do this under these time constraints without sacrificing a minimum level of quality, specifically because we're not treated like an actual profession with the clout or the responsibility that comes with such a thing.

We don't prioritize code over comments because it can be automatically tested, we prioritize code over comments/spec because the code is what gets executed.

And I can say "this can't be done under these time constraints" at my job. If you can't, then you're not being taken seriously because of some other reason, not because you're a programmer.

If that's the way you work, then your employer has made a mistake.

My previous comment ("If that's the way you work, then your employer has made a mistake") was ambiguous, and actually illustrates a point regarding comments (and documentation in general). Like that comment, the short verb phrases that make up source code cannot always express the intent of the programmer (extreme 'self-documenting code' advocates think otherwise, but their experience with simple programs and algorithms doesn't generalize.) In such cases, comments are a good way to explain a design issue that is not obvious.

Who cares about these things, which don't change the code? Anyone who needs to understand the code, which includes anyone working on it, including its original author - writing a brief explanation of a tricky point can help you spot mistakes before you compile or test, or even before you write the code. The goal is to produce correct code, not just code, and the one thing that is faster than continuous testing is avoiding mistakes in the first place.

FWIW, I think the idea that comments need to be complete sentences is silly. Sentences can be as wrong, ambiguous, misleading or uninformative as simple phrases.

I am curious as to how you think this should inform the way we program. You could also say 'coding errors are harmful to the usefulness of programs', but that gives us essentially no insight into how to do things better.

You need code to run code.

You don't need comments to run code.

Do you realize that this observation does not answer the question? There are multiple possible suppositions as to what, if anything, you are implying, so why don't you stop beating about the bush and state clearly the message you are trying to convey?

Your style of commenting is not what this site is about. Go back to Reddit, I'm done.

Literate comments are as a gateway drug to literate programming.

True. However, how useful literate programming is depends on a number of factors, time often being the main limiting factor.

I've found tooling is the main limiting factor. Time certainly doesn't help, but seeing as how I couldn't use literate programming while my colleagues use their IDEs is ultimately what means I will be ditching it once I get help on things.

Yes, that too.

Speaking of which: any good tools for literate programming?

Emacs org-mode has been blowing me away lately. Don't know how well it can scale up to a large project, sadly.

If you're writing code that can be so easily understood that it does not require explanation, get a machine to do it.

A developer I used to know put it thus: "A comment is an apology." I never asked him to explain that, but I guess he meant it's an apology that the code itself is not self-explanatory.

Clear, even verbose commentary can be helpful. But where it often breaks down is when the code is revisited, and the comments are not updated as well. So my practice now is to use comments in early development to help me remember what I was thinking in that bit of code, but then remove most of them during clean-up and refactoring unless something particularly non-obvious is going on. It makes future maintenance of the code less likely to diverge from the (usually unmaintained) comments.

Good code review could help here: "Hey, the diff shows you changed six lines of code in that function but the comments are the same, do they need to be revised?" But I've never worked anywhere where EVERY commit was reviewed, rather a sampling is selected for weekly review. that leaves a lot of opportunity for things to get missed.

Bottom line: extensive comments can add to workload when fixing bugs or making enhancements. So only use them when they really count.

Meh, I'm a huge fan of "readable code" that is easy to follow and just makes sense with no need for explanations. And then there is code that is non trivial.

Too often I've found that vast amounts of effort were put into keeping something "readable" to the point that all of the extra readable hints either just got too large and unwieldy, or it prevented the developer from picking an easy and straight forward solution. (And to be fair, often that developer was me.)

I think there is a very real divide between what the logical flow for a computer to be presented something and the narrative flow for explaining that to another developer are. And no amount of clever variable/function names will really help here.

In fact, often times it can be worse. The proverbial isle of a grocery store. Too much to look at with no obvious "main" method. There is a reason it is a "rookie mistake" to have giant methods that do too much. When it comes down to it many people were only trying to get their program to do one thing. Cluttering their memory with the myriad steps necessary to get there is a large hurdle to get back into something.

The programmer who doesn't change the documentation when changing the code is usually the same one who doesn't change the function name when changing its functionality. If a bad programmer is tweaking your code, it doesn't really matter if the documentation is in the variable names or in the comments.

Contrast PG's style, which had a deep influence on me: https://github.com/arclanguage/anarki/blob/master/arc.arc. There's something to be said for an aesthetic of conciseness that lets one spend less time scrolling around and more time just thinking about what's in one's field of view. That includes stripping comments of unnecessary verbiage.

The article does make one good point: that when you add to a terse comment you often have to add capitals and punctuation. But that is also often a signal to me to ask why I need to switch from communicating a fact to creating a narrative. Is the narrative really necessary? Often I find some way to reorganize the code without needing to narrate it. Relevant is http://steve-yegge.blogspot.com/2008/02/portrait-of-n00b.htm.... (But I'm not claiming everybody who uses complete sentences is a noob.)

I like detailed comments myself, but this argument is hurt by taking them out of their context - it would have been more instructive to examine different styles of commenting to document a complex function or how they can help or hinder structural understanding.

I usually start by writing comments first in plain English based on a variation of psuedocode. That way I'm focused on process first without distraction. Comments are the plan. Clearly articulated steps like any good cookbook. Easy to translate into any language.

// 1. Write plain English psuedocode.

// 2. Translate into scary looking code blocks.

// 3. If maintaining, remember to update comments first (goto Step 1, then Step 2).

The irony is another developer once claimed that my code was generated by a machine. Apparently it was too methodical to be human. Like that was some excuse for him to write crappy code. Urgh! To be great at anything, you need to have excellent communication skills!

Comments should be meaningful. Whether they are sentences or a Morse code is secondary.

I love the attention to detail Ned.

Reply on news.ycombinator.com