GNU Make 4.2 | Hacker News Reader

GNU Make 4.2

lists.gnu.org

80 pointslelf17 comments

One misfeature I discovered recently and I wished got fixed is .SECONDARY vs .PRECIOUS [1]. Both are used to tell Make to keep intermediate products of chains of rules. However, .SECONDARY does not accept wildcards (%.o patterns for example), only .PRECIOUS does. However, .PRECIOUS also adds the guard that an incomplete intermediate product is kept in case Make is interrupted, which you would only ever want for a trivial stamp file. In any other case, you risk ending up with a runt file that breaks your next build.

So if, like any significant project I know, your build specifies lots of wildcard ("implicit") rules, you're stuck using .PRECIOUS, in which case you can never interrupt the build for risk of leaving it in a broken state.

FWIW, djb's "redo" concept [2] is based centrally around atomicity to avoid such a problem.

[1] https://www.gnu.org/software/make/manual/html_node/Special-T...

[2] https://github.com/apenwarr/redo

You get the same problem whenever you build up things piece by piece. So most often I work around this by atomically mv'ing a temporary file to the final product.

    %.foo : %.bar
        echo blubb >$@.tmp
        cat $^ >$@.tmp
        mv $@.tmp $@

If you find yourself using make for data analysis workflows, consider snakemake[0], like make but with the ability to use python and R directly in the rules.

[0] https://bitbucket.org/snakemake/snakemake/wiki/Home

Nice, I've been on the lookout for a good make-for-data-analysis ever since I saw Drake, fell in love with its feature set, gave it a spin, and discovered that it took several seconds just to start up! I would love to get away from my make-based workflow, hopefully this is the one that does it :)

Awesome! How long before this comes to darwin? Is the native version stuck on 3.81 forever?

The $(file ...) function seems like it could be powerful. Are people using that for dynamic dependencies, without needing to use includes?

3.81 is the last release under GPLv2. Apple doesn't ship GPLv3 code.

https://savannah.gnu.org/forum/forum.php?forum_id=4932

Oh, that's why. Godamnit.

Welp, MacPorts it is, then.

Yup. Same reason bash on OS X is an ancient version. And Emacs.

> Apple doesn't ship GPLv3 code.

I wish Apple would stop being so pig-headed about this. They're about the only major tech company that refuses to touch GPLv3. Google, Microsoft, Facebook; they've all got some contact with GPLv3 code without having their empires crumble. In the meantime, Apple's refusal to ship GPLv3 is doing nothing but making it more inconvenient for its customers to get modern software.

But ever since Steve Jobs got pwned by GPLv2 and thus gcc got an ObjC frontend, Apple has been vowing, "nevar forget".

I'm not sure pwned is the right word:

(1) The ObjC/GCC thing happened in the '80s, and didn't prevent them from doing things like shipping (and updating) plenty of GPLv2 software with OS X when it first came out in the late '90s/early 2000s. Additionally, WebKit came out in 2005, and was forked from KHTML, which was LGPL. They didn't have to choose KHTML, they could've written something from scratch or used something else. Their issue is obviously with GPLv3, not the GPL in general.

(2) Many, many, many people, including many people in what you would consider the hacker community, have serious issues with GPLv3. One might even say that for those people, GPLv3 is "doing nothing but making it more inconvenient for its customers to get modern software." And it's not like GPL < v3 was a universally beloved and accepted license either, it's pretty much been a font of endless flamewars since its inception decades ago. I personally think it's a great license as long as developers and users understand its requirements, but let's not kid ourselves that it's something everyone should universally embrace and use regardless of their own goals and ideals, or without fully understanding what requirements it imposes.

(3) Back in the '80s, Jobs asked RMS if the ObjC front end could be distributed separately and linked by the user, and therefore not fall under the linkage requirement of the GPL. RMS said he'd ask his lawyers. He did, and his lawyers said it was a gray area that could potentially be abused. RMS decided to tell Jobs that he would not be OK with that arrangement. Jobs said "OK", et voila. GPL'd ObjC front end. That sounds more like grownups resolving a disagreement than pwnage, but whatever.

I think the GPLv3 is more widely feared than understood. Most of the objections raised in the flamewars come from misunderstandings of its terms (e.g. the belief that it forbids encryption or it has patent anti-retaliation clauses worse than the Apache license), or from objections to early drafts of it. For example, the Linux devs refused to upgrade to GPLv3 after reading an early draft, and never reconsidered the question after the final draft was out.

And yeah, I would consider what happened to Jobs to be "pwning". He was trying very hard to not have to give gcc any source code, and he failed. If he was happy with the GPL, we would have seen more contributions to gcc after that, but they never happened again.

The GPL protects users; if you use software, a GPL license protects your right to make changes to the software via the source code to correct problems and make it suit your needs. I understand that many developers and suits to not wish to give this right to users, but trying to never mention that and calling yourself or anybody "pwned" by the GPL that other developers granted to you is quite simply propaganda.

Protecting users' rights is not getting pwned, it's a failure to pwn your users with software that you don't have pwnership of.

I guess it would be too much trouble for them to switch to BSD make?

Apparently there is no BSD grep so I guess they are stuck with GNU on that one.

> I guess it would be too much trouble for them to switch to BSD make?

Switching to FreeBSD's make would be super simple, but unfortunately it lacks plenty of features that GNU make has.

> Apparently there is no BSD grep [...]

Apple ships FreeBSD's grep.

Yes my mistake; FreeBSD's web man pages are still showing GNU grep even for -current, but my OS X system has BSD grep.

It's a tragedy that the changelog isn't titled "What's GNU:".

Oh come on people. That would be an excellent pun if the word "gnu" followed usual English pronunciation.

Reply on news.ycombinator.com