I always find surprising to find out comments like that regarding code comment. I can understand that someone argues about that writing comments on the code is boring, or that you forget about it or whatever. But to say that the code shouldn’t be commented at all looks a little dangerous to me.
That doesn’t mean that you’ll have to comment everything. Or that adding a comment it’s an excuse to not be clear directly on the code, or the comment should be repeat what is on the code. You’ll have to keep a balance, and I agree that it’s something difficult and everyone can have their opinion about when to comment and when not.
Also, each language has it’s own “comment flow”, and definitively you’ll make more comments on low level languages like C than in a higher level language like Python, as the language it’s more descriptive and readable. Ohhh, you have to comment so many things in C if you want to be able to understand what a function does it in less that a couple of days… (the declaration of variables, for example) #
As everyone has their own style when it comes to commenting, I’m going to describe some of my personal habits commenting the code to open the discussion and compare with your opinions (and some example Python code):
- I put comments summarizing code blocks. That way, when I have to localize a specific section of the code, I can go faster reading the comments and ignoring the code until getting to the relevant part. I also tend to mark those blocks with newlines.
# Obtain the list of elements from the DB
.... [some lines of code]
# Filter and aggregate the list to obtain the statistics
... [some lines of code]
UPDATED: Some clarification here, as I think that probably I have choose the wrong example. Of course, if blocks of code gets more than a few lines and/or are used in more than one place, will need a function (and a function should ALWAYS get a docstring/comment/whatever) . But some times, I think that a function is not needed, but a clarification is good to know quickly what that code is about. The original example will remain to show my disgrace, but maybe this other example (I have copy-paste some code I am working right now and change a couple of things)
It’s probably not the most clean code in the world, and that’s why I have to comment it. Latter on, maybe I will refactor it (or not, depending on the time).
# Some code obtaining elements from a web request ....
# Delete existing layers and requisites
update = Update.all().filter(Update.item == update).one()
UpdateLayer.all().filter(UpdateLayer.update_id == update.item_id).delete()
ItemRequisite.all().filter(ItemRequisite.item == update).delete()
# Create the new ones
for key, value in request.params.items():
if key == 'layers':
slayer = Layer.all().filter(Layer.layer_number == int(value)).one()
new_up_lay = UpdateLayer(update=update, layer=slayer)
if key == 'requisites':
req = ShopItem.all().filter(ShopItem.internal_name == value).one()
new_req = ShopItemRequisite(item=update, requisite=req)
- I describe briefly every non-trivial operation, specially mathematical properties or “clever tricks”. Optimization features usually needs some extra description telling why a particular technique is used (and how it’s used).
# Store found primes to increase performance through memoization
# Also, store first primes
found_primes = [2,3]
''' Find recursively if the number is a prime. Returns True or False'''
# Check on memoized results
if number in found_primes:
# By definition, 1 is not prime
if number == 1:
# Any even number is not prime (except 2, checked before)
if number % 2 == 0:
# Divide the number between all their lower prime numbers (excluding 2)
# Use this function recursively
lower_primes = (i for i in xrange(3,number,2) if prime(i))
if any(p for p in lower_primes if number % p == 0) :
# The number is not divisible, it's a prime number
# Store to memoize
(Dealing with prime numbers is something that deserves lots of comments!) EDIT: As stated by Álvaro, 1 is not prime. Code updated.
- I put TODOs, caveats and any indication of further work, planned or possible.
# TODO: Change the hardcoded IP with a dynamic import from the config file on production.
# TODO: The decision about which one to use is based only on getting the shorter one. Maybe a more complex algorithm has to be implemented?
# Careful here! We are assuming that the DB is MySQL. If not, this code will probably not work.
UPDATE: That is probably also related to the tools I use. S.Lott talks about Sphinx notations, which is even better. I use Eclipse to evelop, which takes automatically any “TODO” on the code and make a list with them. I find myself more and more using “ack-grep” for that, curiously…
- I try to comment structures as soon as they have more than a couple of elements. For example, in Python I make extensive use of lists/dictionaries to initialize static parameters in table-like format, so use a comment as header to describe the elements.
# Init params in format: param_name, value
init_params = (('origin_ip','184.108.40.206'),
for param_name, value in init_params:
- Size of the comment is important, it should be short, but clearness goes first. So, I try to avoid shorting words or using acronyms (unless widely used). Multiline comments are welcome, but I try to avoid them as much as possible.
- Finally, when in doubt, comment. If at any point I have the slightest suspicious that I’m going to spend more than 30 seconds understanding a piece of code, I put a comment. I can always remove it later the next time I read that code and see that is clear enough (which I do a lot of times). Being both bad, I prefer one non-necessary comment than lacking one necessary one.
- I think I tend to comment sightly more than other fellow programmers. That’s just a particular, completely unmeasured impression.
What are your ideas about the use of comments?
UPDATE: Wow, I have a reference on S.Lott blog, a REALLY good blog that every developer should follow. That’s an honor, even if he disagrees with me on half the post
On one of my first projects on C, we follow a quality standard that requires us that 30% of the code lines (not blank ones) should be comments.