From there it should be in the post section (see mate.mk for a simple example)

${SETENV} ${CONFIGURE_ENV} before ${MAKE_CMD} ?

incldue -> include

Where is this used?

I feel those should be prefixed, like UBOOT_MODEL, UBOOT_BOARD_CONFIG...

It feels really strange having a USES that defines PORTNAME, PORTVERSION and CATEGORIES.

THIS SHOULD NOT BE SET.

This is also very wrong.

OK.

Why? What benefit is there from the longer names.

I'ts a typo for UBOOT_OMAP_VERSION. It works because the default version is the same and it picks that up.

Why not?

Why's that?

OK. Will do.

I don't think it's needed, but sure, if that's convention.

namespace polution, all USES that use external variables have them.

All variables that start with an _ are for internal, framework only, use, they should not be used outside.

Well, mostly because it is a noop, it is defined as:

DISTVERSIONFULL= ${DISTVERSIONPREFIX}${DISTVERSION:C/:(.)/\1/g}${DISTVERSIONSUFFIX}

So setting it to something else will not do anything.

This still will not do anything.

Ok, ok, I get what the problem is.

You can't set PORTVERSION in a USES, it is too late.
I'm wondering what other breakage is lurking around as PORTNAME is set too late too.

This *really* should be a master/slave scheme.

Master / slave doesn't work for this. It's more like when we used to build different parts of X11 that was in one big distribution. We didn't do master / slave for that. I couldn't get master / slave working because different revisions needed diferent patches (though maybe github solution would fix that). Other items just were a pain in the backside are easy here. This isn't quite like other things where we have 'variant' ports. Each port is its own thing and only kinda related to other ports of a similar name. That's not the master / slave data pattern.

And the line still actually does something, verified by commenting it out and adding it back.

Why can't I set PORTVERSION in a USES? Please help me understand.

I'll consider this.

Yes, it is a bit strange, but master port route simply didn't work.

It's broken without setting it. If I don't set it, the tar that's grabbed has no version it in:

Attempting to fetch https://codeload.github.com/bsdimp/u-boot/tar.gz/ports-v2016.05?dummy=/u-boot-_GH0.tar.gz

While setting it and DISTVERSIONFULL gives me:

Attempting to fetch https://codeload.github.com/bsdimp/u-boot/tar.gz/ports-v2016.05?dummy=/bsdimp-u-boot-beaglebone-ports-v2016.05_GH0.tar.gz

And yet setting gives me the proper name.

Setting it gives me bsdimp-u-boot-<boardname>-ports-v2016.05_GH0.tar.gz and not setting it gives me two dashes in a row. Hardly a nop.

So
Attempting to fetch https://codeload.github.com/bsdimp/u-boot/tar.gz/ports-v2016.05?dummy=/bsdimp-u-boot-beaglebone-ports-v2016.05_GH0.tar.gz

vs

Attempting to fetch https://codeload.github.com/bsdimp/u-boot/tar.gz/ports-v2016.05?dummy=/bsdimp-u-boot--ports-v2016.05_GH0.tar.gz

Though to be honest, I suspect that I may need to grab a tag, not a branch here. But that's a minor issue.

Ah, it's needed if we do the split.

You could remove the UBOOT_ prefix now.

Please, move that at the top of the file, also move the COMMENT paragraph between PKGNAMESUFFIX and USES, and add a MAINTAINER line just above the comment :-)

This should go just below the PORTVERSION line.

This block should go after the maintainer/comment block

If those can be overridden in slave ports, I think they should be wrapped in .ifs

This feels wrong, it's going to mess up some calculations, you should remove ${STAGEDIR}/ from it.

Mostly I'm looking for a shorthand for where to install stuff, but there's only one or two places so being explicit about STAGEDIR is fine there.

The goal is for them not to be overridden by the slave ports, so I'm starting out w/o the ifs.

All this should go below, say, the MAKE_ARGS line.

I have no problem with shorthands, just don't call it DATADIR, it is a special variable name that the ports framework uses for "things" :-)

OK, but we use these variables in the next few lines... Is that kosher?

make(1) does lazy evaluation, that is, if you do FOO=${BAR} it won't care about BAR until FOO needs to be evaluated.