Re: A critique of cgi.escape

On Mon, 25 Sep 2006 16:48:03 +0200, Max M wrote:

Any change in Python that has these consequences will rightfully be
considered a bug. So what you are suggesting is to knowingly introduce a
bug in the standard library!


It isn't like there have never been backwards _in_compatible changes to
the standard library before.

Ten seconds of googling finds
http://www.python.org/download/releases/2.3/highlights/:

int() - this can now return a long when converting a string with many
digits, rather than raising OverflowError. (New in 2.3a2: issues a
FutureWarning when sign-folding an unsigned hex or octal literal.)

Bastion and rexec - these modules are disabled, because they aren't
safe in Python 2.3 (nor in Python 2.2). (New in 2.3a2.)

Hex/oct literals prefixed with a minus sign were handled
inconsistently. This has been fixed in accordance with PEP 237. (New
in 2.3a2.)

Passing a float to C functions expecting an integer now issues a
DeprecationWarning; in the future this will become a TypeError. (New
in 2.3a2.)

None - assignment to variables or attributes named None will now
trigger a warning. In the future, None may become a keyword.

And more, all from one release.

If the behaviour of cgi.escape is "broken", or incomplete, or misleading,
then Python has a great mechanism for introducing incompatible changes
slowly: warnings.

It isn't good enough to say that the function does what it says it does,
if what it does is dangerous and misleading. Artificial example:

def sqr(x):
"""Returns the square of almost all numbers."""
if x != 1: return x**2
else: return -1

The function does exactly what it says, and yet still has badly dangerous
behaviour that risks introducing serious bugs. If people are relying on
unit tests which include specific tests for that behaviour, then the
function and the code needs to be fixed in parallel. That's what the
warnings module is for.

So any arguments about "breaking code" are a red herring: if cgi.escape
does the wrong thing (and that's arguable), and code relies on that
behaviour, then the code is already broken and needs to be fixed in
parallel with the function. So can we accept that:

(1) *if* there is a problem with cgi.escape it needs to be fixed;

(and, dear gods, I would hope that nobody here wants to argue that Python
should make backwards compatibility a higher virtue than correctness!)

(2) it doesn't need to be fixed *immediately* without warning;

(3) but it can be fixed through a gradual process with warning; and

(4) unit tests and code that expect the (presumed) bad behaviour can be
fixed gradually?

Now that we've got that out of the way, can we CALMLY and RATIONALLY
discuss whether cgi.escape is or isn't broken?

Or, more specifically, UNDER WHAT CIRCUMSTANCES it does the wrong thing?



--
Steven D'Aprano

.

Relevant Pages

  • Re: Beta anti spyware
    ... The irony here is that I've installed the beta on three computers here. ... I've had only one 'bug' that wasn't the fault of the application. ... As I read in here, I don't post often, I find complaints and that's about ... The warning is clearly there in the ...
    (microsoft.public.security)
  • Re: cciss: WARNING/BUG in do_cciss_intr (its back)
    ... Subject: cciss: WARNING/BUG in do_cciss_intr ... I think this is a different bug than the one you reported previously. ... The first warning that now triggers is: ... still using kexec for loading/testing the new kernel. ...
    (Linux-Kernel)
  • Warning: Possible Bug in BIOS DELL Latitude D400_A06 !
    ... Serious Warning - Possible Bug in BIOS update! ... that during both flashing operations no error or ... with DELL that the flashing operation had not been possible from XP. ...
    (Debian-User)
  • Re: cciss: WARNING/BUG in do_cciss_intr (its back)
    ... Subject: cciss: WARNING/BUG in do_cciss_intr ... I think this is a different bug than the one you reported previously. ... The first warning that now triggers is: ... PCI Interrupt Link enabled at IRQ 54 cciss ...
    (Linux-Kernel)
  • Re: huge gcc 4.1.{0,1} __weak problem
    ... We currently give a #warning for 4.1.0. ... The huge problem is that "empty __weak function in the same file and ... several new usages added during this merge window alone. ... least it was when I encountered a gcc bug with these symptoms last year ...
    (Linux-Kernel)