The issue of having an :init pragma for subs came up again during
today's #parrotsketch meeting, and I volunteered to summarize the
discussion for ticket #39926 and the mailing list.
The :init pragma is intended to make it easier for automatically
generated modules to make sure that initializations are performed
prior to execution of :main.
By way of review -- the :main pragma designates the entry point for
a module when it is given to parrot on the command line. If the
module contains multiple subs marked :main, then the last such sub
is used.
However, sometimes the module given to parrot on the command
line actually comes from several sources, such as when using
pbcmerge, or when a compiler includes automatically generated
files coming from pgc or tgc. These external sources may need
to have various initializations performed prior to executing
:main. The easy way to do this is to allow such external sources
to create subs marked :init that are to be executed prior to
:main. Without something like :init, it then becomes the
responsibility of the :main sub to explicitly call any
needed initialization subs, and given that some of these
initialization subs may be coming from other sources, it's
often difficult to know what should be called.
It's easy to confuse :init with the existing :load pragma,
but as :load is currently written it only executes subs
when a module is loaded via the load_bytecode op. Subs
marked :load are not loaded when a .pbc/.pir file comes
from the parrot command line. While we could conceivably
add the "execute before :main" behavior to the :load
pragma, there may be times when we want to differentiate
"execute as program" from "load as library module", so
having a separate :init pragma would allow us to do that.
Also, previous messages in RT #39926 had indicated
that :init would also designate subs to be executed
when loaded as a module, but today we felt that complete
orthogonality might be better. A sub that needs to be
executed in both situations can simply specify both
pragmas (":init :load").
In ticket #39926 both Allison and Chip have already
given their approval for an :init pragma, so this message
is just a summary of where things stand.
More importantly, it's an invitation for people to contribute
tests and code to bring the :init pragma to life. I know that
several of the compiler authors will greatly benefit from
having it available (which is why it came up in today's
#parrotsketch).
Thanks!
Pm
