[Mono-devel-list] MCS initialization cost

[Paolo Molaro]

On 06/23/04 Joe Ante wrote:
> I am using mcs from an application that embeds the mono runtime.
> The startup cost of mcs seems to be pretty high. Around 1.5 second on my
> tiBook. The startup cost is sort of annoying to me because I am used to
> python as a scripting language where I just save and see the result of
> change in code on screen in less than .2 seconds.

On my G4 800 MHz it takes about 1.2 seconds to compile a small program.
Almost all of it is JIT time. If the stuff you usually compile is small,
running mcs with:
	mono -O=-all sources...
will help a bit: the same program compiles in 0.95 seconds. Of course,
with large programs -O=-all is a bad idea.
Of the .95 seconds, 0.85 are still of JIT time.

> Is there any way to reduce it?
> Embedding mcs into a C++ project?
> Adding a new option to mcs which keeps it running and keep on reading the
> standard input? 
> Are there any plans to solve this?

AOT compilation will reduce the overhead a lot, to at least < 0.5
seconds, but it's not yet tested/supported on ppc: some people are
interested though, so expect news on that front in the next few weeks.
We'e looking at improving JIT compile time, too.
A while ago some work was done to make mcs a library you could use, but
I don't know the status of that.

lupus

-- 
-----------------------------------------------------------------
lupus at debian.org                                     debian/rules
lupus at ximian.com                             Monkeys do it better