21 Jun
2024
21 Jun
'24
4:35 a.m.
I defer to your greater experience than mine. I guess I've run into
./configure && make in very vanilla situations, a few Gnu or Gnu-ish
applications. If it has times when it doesn't work, or doesn't behave
well, then I don't doubt your experience.
I first entered this thread in pointing out some similarities in the
style of opaque artificially pseudo intelligence behind systemd and
CMake, namely, don't you decide what to do, tell about these qualities
of these modules and we will decide what to do, don't worry your newbie
little head. I think autoconf and configure are kind of halfway between
user-decides-what to do (straight make) and user-decides-nothing,
is-kept-in the-dark (CMake). So in that way, it's only half as bad. If
it falls over sometimes when it shouldn't I think you know more about
that then me. I will be wary.
For my own code, I stick with straight make, and the occasional script.
On 06/20/2024 02:00 PM, ron minnich wrote:
here's where I'd have to disagree:
"./configure && make is not so bad,
it's not irrational, sometimes it's overkill, but it works"
because it doesn't. Work, that is. At least, for me.
I'm amazed: the autoreconf step on slurm permanently broke the build,
such that I am having to do a full git reset --hard and clean and
start over.
Even simple things fail with autoconf: see the sad story here, of how
I pulled down a few simple programs, and they all fail to build. I
replaced them ALL with a single Go program that was smaller, by far,
than a single one of the configure scripts.
https://docs.google.com/presentation/d/1d0yK7g-J6oITgE-B_odadSw3nlBWGbMK7cl…
On Thu, Jun 20, 2024 at 1:35 PM Luther Johnson
<luther.johnson(a)makerlisp.com <mailto:luther.johnson@makerlisp.com>>
wrote:
I agree that there are certainly times when CMake's leverage has
solved problems for people. My most visceral reactions were mostly
based on cases where no tool like CMake was really required at
all, but CMake had wormed its way into the consciousness of new
programmers who never learned make, and thought CMake was doing
them a great service. Bugged the hell out of me, this dumbing-down
of the general programming population. My bad experiences were all
as a consultant to teams that needed a lot of expert help, when
they had thrown CMake along with a lot of other unnecessary
complexity into their half-working solutions. So I guess it was
all tarred by the same flavor of badly conceived work. But then as
I tried to make my peace with the CMake build as it was, I got a
deeper understanding of how intrinsically irrational CMake is (and
again, behavior changing on the same builds depending on CMake
release versions.
So there certainly are times when something a little more
comprehensive, outside of make, is required. ./configure && make
is not so bad, it's not irrational, sometimes it's overkill, but
it works ... but only if the system is kind of Unix-y. If not you
may wind up doing a lot of work to pretend it's more Unix-y, so
instead of porting your software, you're porting it to a common
Unix-like subset, then emulating that Unix-like subset on your
platform, both ends against the middle. That can be ultimately
counter-productive too.
I have an emotional reaction when I see the porting problem become
transformed into adherence to the "one true way", be it Unix, or
one build system or another. Because you're now just re-casting
the problem into acceptance of that other tool or OS core as the
way it should be. Instead of getting your thing to work on the
other platform, by translating from what your application wants,
into how to do it on whatever system, you're changing your
application to be more like what the "one true system" wants to
see. You've given up control of your idea of your app's core OS
requirements, you've decided to "just give in and be UNiX (or
Windows, or whatever)". To me, that's backwards.
On 06/20/2024 12:59 PM, Warner Losh wrote:
For me, precomputing an environment is the
same as a wysiwyg
editor: what you see is all you get. If it works for you, and the
environment that's inferred from predefined CPP symbols is
correct, then it's an easy solution. When it's not, and for me it
often wasn't, it's nothing but pain and suffering and saying MF
all the time (also not Make File).... I was serious when I've
said I've had more positive cmake experiences (which haven't been
all that impressive: I'm more impressed with meson in this space,
for example) than I ever had with IMakefiles, imake, xmkmf,
etc... But It's also clear that different people have lived
through different hassles, and I respect that...
I've noticed too that we're relatively homogeneous these days:
Everybody is a Linux box or Windows Box or MacOS, except for a
few weird people on the fringes (like me). It's a lot easier to
get things right enough w/o autotools, scons, meson, etc than it
was in The Bad Old Days of the Unix Wars and the Innovation
Famine that followed from the late 80s to the mid 2000s.... In
that environment, there's one of two reactions: Test Everything
or Least Common Denominator. And we've seen both represented in
this thread. As well as the 'There's so few environments, can't
you precompute them all?' sentiment from newbies that never
bloodied their knuckles with some of the less like Research Unix
machines out there like AIX and HP/UX... Or worse, Eunice...
Warner
On Thu, Jun 20, 2024 at 12:42 PM Adam Thornton
<athornton(a)gmail.com <mailto:athornton@gmail.com>> wrote:
Someone clearly never used imake...
There's a reason that the xmkmf command ends in the two
letters it does, and I'm never going to believe it's "make
file".
Adam
On Thu, Jun 20, 2024 at 11:34 AM Greg A. Woods
<woods(a)robohack.ca <mailto:woods@robohack.ca>> wrote:
At Thu, 20 Jun 2024 01:01:01 -0400, Scot Jenkins via TUHS
<tuhs(a)tuhs.org <mailto:tuhs@tuhs.org>> wrote:
Subject: [TUHS] Re: Version 256 of systemd boasts '42%
less Unix philosophy' The Register
cmake (or meson,
"Greg A. Woods" <woods(a)robohack.ca
<mailto:woods@robohack.ca>> wrote:
> I will not ever allow cmake to run, or even exist, on
the
machines I
control...
I'm not a fan of cmake either.
How do you deal with software that only builds with scons, ... whatever the developer decided to use
as the
build tool)?
What alternatives exist short of reimplementing
the
build process in
a standard makefile by hand, which is obviously
very
time consuming,
error prone, and will probably break the next
time you
want to update
a given package?
The alternative _is_ to reimplement the build process.
For example, see:
https://github.com/robohack/yajl/
This example is a far more comprehensive rewrite than is
usually
necessary as I wanted a complete and portable example
that could be used
as the basis for further projects.
An example of a much simpler reimplementation:
http://cvsweb.NetBSD.org/bsdweb.cgi/src/external/mit/ctwm/bin/ctwm/Makefile…
--
Greg A. Woods
<gwoods(a)acm.org <mailto:gwoods@acm.org>>
Kelowna, BC +1 250 762-7675 RoboHack
<woods(a)robohack.ca <mailto:woods@robohack.ca>>
Planix, Inc. <woods(a)planix.com <mailto:woods@planix.com>>
Avoncote Farms <woods(a)avoncote.ca
<mailto:woods@avoncote.ca>>