spkg-configure.m4 for boost and boost_cropped

[dimpase]

we use a macro from autoconf archive to test the version

Depends on #29053

CC: @mkoeppe @isuruf @embray @orlitzky

Component: build: configure

Author: Dima Pasechnik

Branch/Commit: 6f04521

Reviewer: Isuru Fernando

Issue created by migration from https://trac.sagemath.org/ticket/29100

[dimpase]

Dependencies: #29053

[dimpase]

Changed branch from u/dimpase/packages/boostconfig.m4 to u/dimpase/packages/boostconfig

[dimpase]

comment:1

some boost-related package names are missing on #29053, added here

---

New commits:

6f04521

add package names to distros/

[dimpase]

Changed commit from 931cdf8 to 6f04521

[isuruf]

Reviewer: Isuru Fernando

[orlitzky]

comment:6

Works for me on Gentoo but it's a little scary to have the check for boost implicitly be the check for boost_cropped. It would be nice if the new macro result was cached -- then we could just repeat the check -- but I don't think it is.

Maybe check HAVE_BOOST as a safety net against changes to boost_cropped? I dunno.

[dimpase]

comment:7

No distro I tried allowed for
an incomplete install of boost.

I guess it is a historical artefact that Sage has these two packages

[jhpalmieri]

comment:8

Replying to @dimpase:

No distro I tried allowed for
an incomplete install of boost.

I guess it is a historical artefact that Sage has these two packages

Sage needs boost_cropped, not all of boost. The boost tarball is ten times the size, boost_cropped took 8 seconds to build on my machine, and boost took 21 minutes. So if by "historical" you mean before this ticket, I guess "historical arfifact" is accurate.

[dimpase]

comment:9

I don't know how boost_cropped was created, at least spkg-src provides no clue.

I don't know a modern Linux distro that supplies such a "cropped" thing.

[jhpalmieri]

comment:10

According to spkg-src, boost_cropped is obtained from boost by just taking the subdirectory boost (i.e., boost_1_66_0/boost).

I'm almost positive the cropped version is a pure Sage invention, to cut down on size and to speed up building.

[orlitzky]

comment:11

I think I see what happened. The sage library doesn't need boost at all, the boost spkg is optional. But the spkg-configure.m4 for boost causes it to be installed whenever a system copy is not present.

Instead, I think we should leave boost alone (why is it an optional package to begin with?). The spkg-configure.m4 for boost_cropped should look for a full version of boost, like it does now. And then if someone decides to install the optional boost spkg... who cares what happens.

[dimpase]

comment:12

Replying to @orlitzky:

I think I see what happened. The sage library doesn't need boost at all, the boost spkg is optional. But the spkg-configure.m4 for boost causes it to be installed whenever a system copy is not present.

Instead, I think we should leave boost alone (why is it an optional package to begin with?). The spkg-configure.m4 for boost_cropped should look for a full version of boost, like it does now. And then if someone decides to install the optional boost spkg... who cares what happens.

I don't understand what I am supposed to reproduce. I think an optional package only gets installed if its installation is explicitly triggered by make or sage -i, and not merely because ./configure was not able to find a system copy.

[dimpase]

comment:13

we have a number of optional and experimental packages with spkg-configure.m4 already, and nothing like this was ever reported.

[orlitzky]

comment:14

Ok, I'm going to put it back to positive review then. I'm not willing to delete boost at the moment to see what happens.

The check for boost_cropped now looks even more correct to me, since that's the standard package. My comment about boost's spkg-configure.m4 no longer applies, it isn't standard and doesn't need to be detected.

[orlitzky]

comment:15

Sorry, false alarm: I tested with some other packages. If I uninstall the optional gp2c, for example...

./configure
...
checking SPKGs to install... 
    ...
    gmp-6.1.2 will not be installed (configure check)
    gmpy2-2.1.0b1
    gp2c-0.0.10.p0
    ...

That sure looks like gp2c will be installed from the spkg. That message combined with John's comment made me think for a minute that a full copy of boost was being installed in addition to (or in lieu of) boost_cropped.

[mkoeppe]

comment:16

#28788 addresses improving the printing of this message.

[embray]

comment:17

Yikes, that ax_boost_base.m4 is scary. I hope I never have to debug anything in it. But if it works it works, and I imagine (?) it's well tested.

[vbraun]

Changed branch from u/dimpase/packages/boostconfig to 6f04521