[Python-Dev] Re: Update PEP 292

[Tim Peters]

[Brett]
> That is coming down to a question of convenience compared to
> explicitness.  Do we want "Give me $1,000,000 now!" to raise ValueError,
> or do we want to let it pass through?
>
> "Explicit is better than implicit" is good since it will cut down on
> errors (which is what Tim is arguing for along with "Special cases
> aren't special enough to break the rules").
>
> But Raymond is pushing for "practicality beats purity" which is more or
> less true,

It's far less true for non-Americans.  I know it took me several
decades to learn that "$ means currency" isn't an international
standard enforced by the death penalty -- although it should be
<wink>.

> but I am starting to agree with Tim that dollar amounts in an
> interpolated string probably won't hapeen that often.  If anything the
> number will be what is interpolated in.

And that's where "handy shortcuts" bite too.  "OK, I want to
interpolate the value of variable 'dollars'.  A $ sign on its own
means a $ sign, and to interpolate the value of 'dollars' I do
$dollars, so putting them those together I must need

    "Total: $$dollars"

".  Which doesn't do what they want, although

    "Total: $ $dollars"

might be close enough.  A *consistent* (if you want $ in the output,
you must have $$ in the input) rule may be surprising at first, but it
doesn't remain surprising.

> So going for a more explicit version is seeming to be the better solution.
> Perk of the current implementation is that if someone wants a looser definition
> they can just change the regex themselves (you guys could even go as far as to
> break up the regex into individual class attributes and just have a
> property that does a '|'.join on them as a read method so it is even
> easier to just change the one group).

We could indeed make this so complicated nobody will understand it
<wink>.  The use of named (as opposed to the current numbered)
capturing groups could make it a lot less dependent on details of how
the regexp happens to be written, though.

> So I guess it's your call, Mr PEP Champion, on whether you want all
> individual, non-interpolating $ signs to lead to an error or just random
> orphaned ones at the end of the line.

I still don't grasp why the end of the line is somehow special here. 
The real thrust of what I'm after is catching what will *certainly* be
common typos, like

    $(barry)
    ${barry)
    $[barry]
    $(barry}
    ${barry{
    ${barry

The specific gimmick I suggested flags those as errors, and flags an
odd number of trailing $;s too, but the latter isn't a kind of typo I
expect occurs in real life.