On Aug 6, 2006, at 4:08 PM, Leopold Toetsch wrote:
Am Sonntag, 6. August 2006 19:03 schrieb Will Coleda:
we get a 9K line PIR sub. Running this through parrot dies after
about 2.5m on an runtime error on one of the tests.
Great progress. While Parrot shouldn't die or - in the long run -
take really
long to compile that, you could work a bit on the compiler in the
meantime.
I just wanted to make sure I mentioned the error here because I knew
someone would try to run this and get the error: I think there's a
ticket opened for this particular error already...
Here's a snippet from near the end of the code:
start_1559:
dynamic_1559:
.local pmc command
$P1566 = new .String
$P1566 = 'foreach'
# why create a new String here - the line below clearly shows, that
the
compiler wants to call '&foreach'.
This should be moved down to the unknown section below - if the .sub
isn't found at runtime, we need to dispatch the name to the unknown
handler. So this string needs to be here, but it's in the wrong spot.
I'll fix that.
push_eh invalid_1559
command = get_hll_global '&foreach'
# why '&foreach' the tcl name is 'foreach' - just install that sym
in the
runtime namespace and use it. Why the 'get_hll_global'? You are
setting up an
exception frame anyway. Why not just compile to:
Because tcl allows both [foo] and $foo. So we (internally) adopted
the sigils to keep these two things segregated. We could have two
separate namespaces for them, but the thought was that this is a
better choice for interoperability.
push_eh invalid_xyz
$P1565 = 'for_each'($P1560, $P1561, $P1563)
clear_eh
...
I suppose we could have the exception handler check to see if the
exception was due to a missing sub, and unknown() if so, or rethrow
if not. That might work, though I'd be curious if parrot did this
lookup at each invocation or somehow cached it.
invalid_1559:
.local pmc interactive
interactive = get_root_global ['tcl'], '$tcl_interactive'
# you might cache that once - it's probably used in zillion of places
This is a user visible variable that could theoretically change
anywhere, so any caching scheme has to take that into account.
unless interactive goto err_command1559
.local pmc unk
unk=find_global '&unknown'
# same here and just do ...
'unknown'(...)
... call it instead of ...
unk($P1566, $P1560, $P1561, $P1563)
my 2 c
leo
I'll investigate just using the '&foreach'() syntax in both cases,
and there's probably some wins there.
A lot of the compilation at this point is naive - we don't bother to
do optimizations with constant strings, for example: constants,
command interpolation, variable interpolation - all these end up
naively in a PMC register that we can than manipulate uniformly. We
can for sure be more aggressive about this sort of thing, though.
Another thing that I'm considering is a custom op library for this
sort of dispatch. Let it handle things at the C level - which might
not be too much faster/slower at execution time, but would be a plus
during compilation.
We could probably replace the entire section above with:
tcl_dispatch $P1566, $P1560, $P1561, $P1563
and have it do all the heavy lifting.
Thanks for the feedback, I'll incorporate what I can!
--
Will "Coke" Coleda
[EMAIL PROTECTED]
