Philosophical quandary: I’m using 

/bin/cat

as part of a build step. It’s a POSIX utility in its POSIX location, and I’m using POSIX-compliant args (i.e. nothing implementation-specific) for it. Should I use a 

BinaryPath

 request to find that (and be consistent with the rest of Pants), or am I OK directly calling 

/bin/cat

and saving an extra step in the graph?

I think as long as you restrict yourself to POSIX-compliant args, that seems sensible. You get into trouble when using newer features and the user wants

/usr/bin/cat

used over

/bin/cat

etc. This has happened to us with

zip

,

unzip

, and

tar

commands That restriction should be documented in the source code, ideally with a link to the relevant docs

I vote to not worry about it now. It’s easily refactored to use

BinaryPath

should there be a user who has

cat

in a non-standard location.

Great

and if it is a performance problem, we can always write an engine intrinsic to concatenate 2+ files together

in this case, we need to postprocess the output of

cat

anyway, so filesystem operations involved in sandbox operations would probably eclipse the savings of using an engine intrinsic