Hacker News Reader

Comment by vog | Hacker News Reader

If you aim for slightly larger projects and efficiency, you'd need at least a DAG (because that's the fundamental nature of a dependency graph).

You can find a nice explanation in the "tup" build tool, although there are of course many others: http://gittup.org/tup/

And yes, you can express DAGs with text and/or filesystem operations (plus symlinks), but not efficiently and writing code that operates on that level is cumbersome.

View on HN

Regarding DAG: tsort (1) sorts input lines by interpreting them as a list of edges. I guess that's what you where referring to when you said "cumbersome" :)

    $ cat > f
    a c
    a d
    b c
    ^D
    $ tsort f
    a
    b
    d
    c

The problem is not tsort, it is adding, removing, updating items in such a text file. (not to mention nasty corner cases like filenames containing whitespaces, or the security implications of filenames containing newlines)

Tup is pretty darn amazing. I've never had such a high success rate while writing build rules under it. "Once it builds, it has no bug". Not to mention the great speed to begin with.

Is there any similar build system? I know of https://github.com/droundy/fac, but it's not cross-platform

tup sounds similar to qbs. I found qbs to be a most amazing C++ build system, it's the only one I used that is

1) reasonably easy to use 2) very fast 3) very correct

Most build systems fail 3), usually 2) and often also 1).

What makes tup amazing is it traces which files a program touches, so assuming your programs obey the rule "I need to rerun if I have different command line arguments or something I read has changed", you get perfect dependencies for any program. I use it for running experiments with custom programs and scripts all the time, and it just refund the parts I need.

Ah, thanks for the pointer to `tup`! I'd spent a while looking for it a few months ago but couldn't remember the name.

Reply on news.ycombinator.com