[Omake] generating a list of commands to do for cross compiling

[Aleksey Nogin]

On 10.01.2007 15:55, Erick Tryzelaar wrote:

> Say we were cross compiling a program to a platform that doesn't support 
> ocaml, and thus omake. Would it be possible to get omake to print out a 
> list of all the operations that would be done? The "-n" option doesn't 
> work as it errors out if a target doesn't exist:

Eric,

You are right - the -n option appears to be broken currently and 
probably needs to be fixed (more on this below).

Short of posting OCaml and OMake runtimes (just enough of it to get a 
bytecode OMake running), the most obvious options would seem to be:

- Compile on a platform that does support OMake and is sufficiently 
similar. Collect the build steps (possibly throwing away the scanner 
steps, if you are sure that they do not have "important" side-effects). 
Possibly tweak (manually?) some of the command-line arguments to match 
the right platform.

- Set up a proper cross-compiling environment. Run omake normally, 
possibly after adding some support for passing the cross-compilation 
flags to all the compiler invocations.

Now back to the -n option. The current problem with it is easily fixable 
(and ought to be fixed) - basically, as OMake code was being modified 
and improved, we forgot to maintain the support for -n and now several 
places in the code fail to account for a possibility of -n and do the 
right thing.

Still, I am not sure what the -n option should mean. In general, OMake 
can affect things in 3 ways:
1) The evaluation of the OMake code itself may have side-effects (for 
example, you are allowed to spawn external commands from the OMake code, 
not just from the rules, although in most cases this is very bad style).
2) Scanner rules.
3) Regular rules.

While it is obvious that "-n" ought to disable the regular rules, the 
rest is quite unclear. On one hand, the scanner rules ought to be 
disabled to (for one, some scanner rules will in fact partially build 
the target, which we want to avoid; second, the scanner rules may depend 
on the targets that are built using the the regular rules). On the other 
hand, without running the scanner rules, we will not know all the 
dependencies, so the meaning of "-n" as "show me what would happen if I 
run it" will be compromised. Finally, the first category is even less 
clear as it is not even obvious how to decide what belongs in it, it is 
unclear whether there would be a reasonable way of disabling it and 
disabling it is likely to change the outcome in an unpredictable way, 
making the "-n" even less close to what would happen without it.

Aleksey