Re: Guile Startup Time Suggestion

[Andrew Archibald <aarchiba@calum.csclub.uwaterloo.ca>]

Tel <telford@eng.uts.edu.au> writes:

> If the regexp state machine is constant then is there any reason why
> hobbit cannot compile the initialised regexp data into the code. This is
> pretty much what the flex utility does for C code. C programs have supported
> the concept of compiling in initial variable values for quite some time.

Well, yes and no.  C will compute initial variable values if you're
lucky.  But, for example:

int fac(int n) { if (n) return n*fac(n-1); else return 1; }

int k=fac(10000);

This will get computed at runtime, not compile time.  Myabe if you
inlined the function, you could explode the compiler's brain.

Anyway, this is closer to the situation in guile-lang.  Hobbit could
do that, but the C compile still wouldn't fix it up.  And besides,
it's currently computed lazily.  The freezer is worth looking at
(another form of unexec, really), because it can freeze scheme data
into C code.  But here, we have great honking C data structures
(shudder). 

> This is also a good point because you know how many pointers to expect
> in your data and where to expect them. In theory the (write) procedure
> can serialise any non recursive scheme data structure but in practice
> this may not be useful for all situations.

Yes, but scheme data structres aren't really all that problematic.
The big problem is the C structures and SMOBs which ware difficult to
pickle.  And it doesn't pickle vectors, lambdas, or continuations.
Lambdas are definitely useful to have in a sheme data structure. 

Andrew