[Python-3000] Minor addition to Python interactive shell...

[Alex Martelli]

You're welcome!  I wasn't trying to discourage you -- I was trying to
prompt you to do it the best way, as a third party open source project
(perhaps a contributed one to iPython, etc).  Brett suggests that the
index might be already around somewhere (maybe just for the standard
library, but that would be a start) so if you could locate it then the
needed excepthook might not be too hard to write, etc, etc.  However,
it's definitely up to you!

Alex

On Sat, Aug 23, 2008 at 2:45 PM, Abdallah El Guindy
<abdallah.elguindy at gmail.com> wrote:
>> Adding a package to the repositories searched by apt-get is a much
>> higher-ceremony operation than copying a file into some sys.path
>> directory or changing sys.path, so that indexing makes sense for
>> apt-get but not for Python's imports -- I can imagine the poor Python
>> interpreter churning away all the time updating the index all the
>> time.
>
> You are definitely right about that, I totally missed this when first
> thinking about it...
>
>> there is absolutely
>> nothing to be gained by having the functionality inside the core
>> rather than in a third-party package.
>
> I guess you are right once again.
>
>> If and when your extension "takes the Python world by storm"
>
> I guess this is never gonna happen (that's why I probably tagged the feature
> minor).
>
> I did not realized at the beginning that it has as much technical
> complications.. Thanks for pointing that out, I'm convinced that perhaps it
> is not a good idea after all.
>
> Thanks.
>
> On Sun, Aug 24, 2008 at 12:35 AM, Alex Martelli <aleaxit at gmail.com> wrote:
>>
>> On Sat, Aug 23, 2008 at 2:10 PM, Abdallah El Guindy
>> <abdallah.elguindy at gmail.com> wrote:
>> > I believe there must be a way... Maybe by creating an index file for
>> > each
>> > module. I'm not sure, but I think the number of packages on apt-get is
>> > much
>> > more than the number of python built-in modules (obviously I don't know
>> > their number), yet it is doable with the case of apt-get.
>>
>> Adding a package to the repositories searched by apt-get is a much
>> higher-ceremony operation than copying a file into some sys.path
>> directory or changing sys.path, so that indexing makes sense for
>> apt-get but not for Python's imports -- I can imagine the poor Python
>> interpreter churning away all the time updating the index all the
>> time.  And even then, the Python index WOULD still potentially miss
>> some entries, because a Python module is vastly more dynamic than the
>> totally-static repositories used by apt-get, where each package is
>> mandated to very explicitly state what it requires, and what it
>> provides.
>>
>> If you want to try your hand at this kind of extension, I recommend
>> you do it as a pypi package -- since all you want is peculiar messages
>> in response to some NameError or ImportError exceptions, you can hang
>> your functionality neatly into sys.excepthook -- there is absolutely
>> nothing to be gained by having the functionality inside the core
>> rather than in a third-party package.  Moreover, I would recommend
>> targeting the extension (solely or primarily) at "rich" environments
>> such as IDLE or ipython (http://ipython.scipy.org/moin/), whose users
>> already expect and obviously appreciate getting especially rich
>> interactive functionality even if it comes at the potential cost of
>> some overhead -- users of the Python built-in interactive interpreter
>> are more likely to be after lean-and-mean functionality instead.
>>
>> If and when your extension "takes the Python world by storm", you will
>> have strong practical arguments to recommend its inclusion in some
>> future version of Python -- its popularity will have proven that there
>> is strong demand for that functionality, after all.  Until such
>> support does exist, it's hard to argue for including in the Python
>> core offering something that can be done just as well as a 3rd party
>> package, has no proven demand, AND presents potentially delicate
>> tradeoffs in terms of implementation (is the user going to be asked to
>> explicitly reindex each time they alter sys.path, or is the overhead
>> of the reindexing going to be implicit in ANY alteration at all?  etc,
>> etc).
>>
>>
>> Alex
>
>