I don't think that we're going to be able to have "one generator to
rule them all" thanks to use of templates. I do, however, think that
templates on a per-language basis will reduce the amount of generator
code needed and improve readability by a LOT.
-Bryan
On Mar 6, 2009, at 11:28 AM, Mark Slee wrote:
I'd just point out that there are more differences between the
language generators than it seems on the surface. We tried to have
much more code reuse across the language generators than we ended
up with, especially for C++/Java etc. which seem highly similar,
but it turned out in practice to get really awkward really fast, so
we erred on the side of more standalone language generators that
are easier to understand in isolation.
I still think templates with more shared code for different
generators is a good avenue to explore, but just want to raise a
caution flag that it's been tried, and is reasonably difficult.
-----Original Message-----
From: Esteve Fernandez [mailto:[email protected]]
Sent: Friday, March 06, 2009 3:04 AM
To: [email protected]
Subject: Re: Some thoughts about changes to Thrift
On Friday 06 March 2009 03:24:23 Mark Slee wrote:
That's pretty much my fault.
Actually the current Thrift compiler served its puporse quite
decently, but the coming of new languages and features has made it
(along with the
generators) a bit more complex and less easy to maintain.
I personally like Java and think it'd be a decent choice, but
there do
seem to be a decent number of people out there building services who
seem to vehemently hate Java. Lots of *nix systems do not have a JVM
or JDK installed by default -- and it's pretty annoying for users who
aren't writing services in Java to have to install it to build the
compiler.
I'd argue for writing a parser in Java and use a template engine
(Velocity, Freemarker, etc.). Why not Python? Dunno :-) I think
Java is more widely deployed, it has better tools for scanning and
parsing and more developers.
Etch, another incubating project with similar goals to Thrift, uses
JavaCC for parsing and Velocity as template engine for its compiler
and generators, even though it also emits C#
Using a template engine would make development of new features and
the inclusion of more languages much easier. Instead of having to
hack a generator to add some new feature, we could simply supply a
new template file. For example, at some point in the future I'd
like to build a generator for Javascript, and if we used a template
engine, it would be a matter of writing a minimal generator and a
bunch of templates.
Cheers.
