Re: non-recursive build question

[David Boyce]

It's always struck me that the best way to answer the out-of-date question 
generically is via a "plugin" design. Any implementation that's hardwired 
into make is bound to be wrong for someone's needs. Couldn't make do 
something like this: at startup, it uses the dlopen/dlsym() interface 
(there's something similar on Windows) to look for a function with a 
particular name and a documented signature including a boolean return 
value. If found, that's what it uses to make subsequent out-of-date 
decisions. If not (the normal case), it falls back to an internal function 
which simply implements the forward-timestamp-based logic that's always 
been standard.

This would allow for alternate implementations of statefulness, entirely 
configurable by the client site. Developers would only need create a shared 
library implementing the documented function. They could then build gmake 
so as to link it in explicitly, or LD_PRELOAD it on an as-needed basis. I 
could see a whole cottage industry of plugins developing.

One limitation is that GNU make has probably been ported to some old 
platforms which don't have dlopen/dlsym or similar. But IMHO it wouldn't be 
so bad to make this feature depend on a HAVE_DLFCN_H configure macro and 
thus be available only on modern platforms, which all support something 
similar AFAIK. Certainly SUS specifies dlopen.

Would this work or am I misunderstanding something about dlopen?

-dsb