[lsb-discuss] Perl, Python in LSB 3.2

[Wichmann, Mats D]

> It's common, and fairly straight-forward, to generate the
> bytecode files
> at installation time (through the compileall module). Then, you won't
> suffer from the performance hit of dynamic compilation on each program
> startup, and still are independent of the precise version of the
> interpreter installed. 
> 
>>> If the only case supported by the LSB is source-only
>>> modules, then the magic cookie is no issue:
>> The LSB cannot force people to ship source, that would not be the
>> right thing to do.
> 
> It's important to notice that this is an explicit decision to be made
> be the body defining LSB. It is not automatically a given that LSB
> must support shipping byte-code files across systems 

Right, this would be useful to hear comment on.

Is it okay that .pyc (and maybe .pyo?  shudder) files be
generated at install time, rather than included in the
installation packge itself?  That restriction would
certainly make an initial implementation easier. 

Note there has been grumblage elsewhere when there are files
the package manager is not aware of, so one of the tricks
to inform the package manager of these files would
probably need to be employed (this is not the same as
the non-package-manager problem at all; rpm would know
about these and does have a mechanism for letting the
database know; whether that would translate across 
alien to debs I'm not sure)