Is there a need for a "use strict" Python compiler?

Well, I'm not much of a python programmer, but I'd say that the answer is 'YES'.

Any dynamic language that lets you create a variable with any name at any time, could use a 'strict' pragma.

Strict vars (one of the options for strict in Perl, 'use strict' turns them all on at once) in Perl requires that all variables are declared before they are used. Which means that this code:

my $strict_is_good = 'foo';$strict_iS_good .= 'COMPILE TIME FATAL ERROR';

Generates a fatal error at compile time.

I don't know of a way to get Python to reject this code at compile time:

strict_is_good = 'foo';strict_iS_good += 'RUN TIME FATAL ERROR';

You will get a run-time exception that strict_iS_good is undefined. But only when the code is executed. If your test suite does not have 100% coverage, you can easily ship this bug.

Any time I work in a language that does not have this behavior (PHP for example), I get nervous. I am not a perfect typist. A simple, but hard to spot, typo can cause your code to fail in ways that may be hard to track down.

So, to reiterate, YES Python could use a 'strict' pragma to turn on compile time checks for things that can be checked at compile time. I can't think of any other checks to add, but a better Python programmer probably could think of some.

Note I focus on the pragmatic effect of stict vars in Perl, and am glossing over some of the details. If you really want to know all the details see the perldoc for strict.

Update: Responses to some comments

Jason Baker : Static checkers like pylint are useful. But they represent an extra step that can be and often is skipped. Building some basic checks into the compiler guarantees that these checks are performed consistently. If these checks are controllable by a pragma, even the objection relating to the cost of the checks becomes moot.

popcnt : I know that python will generate a run time exception. I said as much. I advocate compile time checking where possible. Please reread the post.

mpeters : No computer analysis of code can find all errors--this amounts to solving the halting problem. Worse, to find typos in assignments, your compiler would need to know your intentions and find places where your intentions differ from your code. This is pretty clearly impossible.

However this does not mean that no checking should be done. If there are classes of problems that are easy to detect, then it makes sense to trap them.

I'm not familiar enough with pylint and pychecker to say what classes of errors they will catch. As I said I am very inexperienced with python.

These static analysis programs are useful. However, I believe that unless they duplicate the capabilities of the compiler, the compiler will always be in a position to "know" more about the program than any static checker could. It seems wasteful not to take advantage of this to reduce errors where possible.

Update 2:

cdleary - In theory, I agree with you, a static analyzer can do any validation that the compiler can. And in the case of Python, it should be enough.

However, if your compiler is complex enough (especially if you have lots of pragmas that change how compilation occurs, or if like Perl, you can run code at compile time), then the static analyzer must approach the complexity of the compiler/interpreter to do the analysis.

Heh, all this talk of complex compilers and running code at compile time shows my Perl background.

My understanding is that Python does not have pragmas and can not run arbitrary code at compile time. So, unless I am wrong or these features are added, a relatively simple parser in the static analyzer should suffice. It certainly would be helpful to force these checks at every execution. Of course, the way I'd do this is with a pragma.

Once you add pragmas to the mix, you have started down a slippery slope and the complexity of you analyzer must grow in proportion to the power and flexibility you provide in your pragmas. If you are not careful, you can wind up like Perl, and then "only python can parse Python," a future I wouldn't want to see.

Maybe a command line switch would be a better way to add forced static analysis ;)

(In no way do intend to impugn Python's capabilities when I say that it can't futz with compile time behavior like Perl can. I have a hunch that this is a carefully considered design decision, and I can see the wisdom in it. Perl's extreme flexibility at compile time is, IMHO, a great strength and a terrible weakness of the language; I see the wisdom in this approach as well.)

Python does have something that can change script syntax:

from __future__ import print_function

and various other future-features that have syntax implications. It's just that Python's syntax has been stricter, stabler and more well-defined than historical Perl; the kind of things that ‘strict refs’ and ‘strict subs’ prohibit have never existed in Python.

‘strict vars’ is primarily intended to stop typoed references and missed-out ‘my’s from creating accidental globals (well, package variables in Perl terms). This can't happen in Python as bare assignments default to local declaration, and bare unassigned symbols result in an exception.

(There is still the case where users accidentally try to write-through to a global without declaring it with a ‘global’ statement, causing either an accidental local or, more often, an UnboundLocalError. This tends to be learned fairly quickly, but it is an arguable case where having to declare your locals could help. Although few experienced Python programmers would accept the readability burden.)

Other language and library changes that do not involve syntax are handled through the warnings system.

"the run-time binding philosophy that Python embraces... makes "use strict" behavior unnecessary [and] especially undesirable"

Pretty good summary. Thanks.

That is essentially it. Static analysis tools don't help Python enough to be worthwhile.

Edit

"I'm asking for us to introspect on why we don't need it and, relatedly, why Perl programmers think they do need it."

The reason why is precisely the reason you already gave. We don't need it because it doesn't help. Clearly, you don't like that answer, but there's not much more to be said. Compile-time or pre-compile time checking simply does not help.

However, since you took the time to asked the question again, I'll provide more evidence for the answer you already gave.

I write Java almost as much as I write Python. Java's static type checking does not prevent any logic problems; it doesn't facilitate meeting performance requirements; it doesn't help meet the use cases. It doesn't even reduce the volume of unit testing.

While static type checking does spot the occasional misuse of a method, you find this out just as quickly in Python. In Python you find it at unit test time because it won't run. Note: I'm not saying wrong types are found with lots of clever unit tests, I'm saying most wrong type issues are found through unhandled exceptions where the thing simply won't run far enough to get to test assertions.

The reason why is Pythonistas don't waste time on static checking is simple. We don't need it. It doesn't offer any value. It's a level of analysis that has no economic benefit. It doesn't make me any more able to solve the real problems that real people are having with their real data.

Look at the most popular SO Python questions that are language (not problem domain or library) related.

Is there any difference between "foo is None" and "foo == None"? -- == vs. is. No static checking can help with this. Also, see Is there a difference between `==` and `is` in Python?

What does ** (double star) and * (star) do for parameters? -- *x gives a list, **x gives a dictionary. If you don't know this, your program dies immediately when you try to do something inappropriate for those types. "What if your program never does anything 'inappropriate'". Then your program works. 'nuff said.

How can I represent an 'Enum' in Python? -- this is a plea for some kind of limited-domain type. A class with class-level values pretty much does that job. "What if someone changes the assignment". Easy to build. Override __set__ to raise an exception. Yes static checking might spot this. No, it doesn't happen in practice that someone gets confused about an enum constant and a variable; and when they do, it's easy to spot at run time. "What if the logic never gets executed". Well, that's poor design and poor unit testing. Throwing a compiler error and putting in wrong logic that's never tested is no better than what happens in a dynamic language when it's never tested.

Generator Expressions vs. List Comprehension -- static checking doesn't help resolve this question.

Why does 1+++2 = 3? -- static checking wouldn't spot this. 1+++2 in C is perfectly legal in spite of all the compiler checking. It's not the same thing in Python as it is in C, but just as legal. And just as confusing.

List of lists changes reflected across sublists unexpectedly -- This is entirely conceptual. Static checking can't help solve this problem either. The Java equivalent would also compile and behave badly.