I suppose changing sh would be a better solution. The only disadvantage there is we need to make sure that anything which would otherwise use system for invoking potentially-problematic executables should instead use sh. Optimally, the JRuby guys should eventually adopt something like this for the canonical system implementation, but until then...
I'll make the necessary changes in my fork. Daniel On Fri, Apr 17, 2009 at 8:17 PM, Assaf Arkin <ar...@intalio.com> wrote: > On Fri, Apr 17, 2009 at 11:42 AM, Daniel Spiewak <djspie...@gmail.com > >wrote: > > > I think we should consider this: > > > > > http://github.com/djspiewak/buildr/commit/51280abe5d37c2800ae53828bc0bb3ba5b51d946 > > > > The reason being that JRuby's Kernel#system method is fundamentally > broken > > (because it is based on java.lang.Runtime.exec). They've managed to > piece > > things together so that it's rarely an issue, but problems do surface > when > > doing things like the following: > > > > system 'scala' > > > > If you run this from JRuby, your terminal will freeze and refuse to > accept > > any input (except for escape sequences like Ctrl+C). This limitation in > > Kernel#system basically prevents any sort of working REPL support (git:// > > github.com/djspiewak/buildr.git / interactive-shell). In order to get > > around this, I propose that Buildr should rewrite the Kernel#system > > method *when > > running under JRuby* so that it uses the platform's native `system` > > function > > (from the C stdlib). The commit linked above (51280ab) contains the > > necessary implementation. Note that this commit is basically untested, > but > > it seems to work on Mac for everything I've tried so far. > > > > Advantages: Buildr's external process dispatch on JRuby will be more > > predictable and basically in line with what happens on MRI. This will > > probably resolve/avoid some bugs and certainly opens up a whole new > series > > of possibilities (like the interactive shell support). > > > > Disadvantages: It introduces a new gem dependency on JRuby (ffi-0.3.2) > and > > actually overwrites a core Ruby method. It's the latter that has me the > > most concerned. Kernel#system is quite foundational to Buildr and Ruby > in > > general, so casually monkey-patching in a new implementation is not a > step > > which should be taken lightly. > > > I also have a bad feeling about changing Kernel#system, but we can > certainly > change sh, which is brought over from Rake. > > Assaf > > > > > > > > Thoughts? > > > > Daniel > > >
