Re: CVS commit: pkgsrc/math/py-Scientific-doc

[Joerg Sonnenberger <joerg%bec.de@localhost>]

On Sun, Jul 31, 2016 at 03:32:50PM -0700, Alistair Crooks wrote:
> These are hardly essential build instructions, since AUTO_MKDIRS being set
> in the PLIST will DTRT. If it doesn't, your PLIST is incorrect.

But that's the crux, isn't it? There are many situations in which the
PLIST is incomplete or otherwise non-existant. I already gave the
example of updating a package. AUTO_MKDIRS does not work correctly in
such situations.

> Logically, AUTO_MKDIRS is the correct solution, and I cannot understand why
> anyone would argue against it

So far the build of a package works correctly with or without the PLIST.
AUTO_MKDIRS is introducing a dependency of the installation phase on the
PLIST, which often is populated only *after* the installation phase.
Think of this feature as a layering violation if you want. For many
packages, the absence of the PIST will trigger a clearly visible build
error. In that case, using AUTO_MKDIRS is harmless. What I am arguing
against is sweeping changes, especially for cases for INSTALLATION_DIRS
contains only one or two entries anyway.

Joerg