makepkg breaks 'make menuconfig' with 2.6.28.1

kludge · 2009-01-21 09:13:23

i'm attempting to compile a custom kernel using the PKGBUILD and instructions from Custom Kernel Compilation with ABS. i'm using 2.6.28.1; in fact, all i want to do is enable ACPI debugging messages. after wrangling a deprecated debugging call, i'm already to roll.

except that when makepkg calls 'make menuconfig' inside the build() function, the menu is unuseable. the arrow keys produce '[[B[[C[[A', etc., and i can't select any of the options. i can only get out of the menuconfig with <ctrl>-C, which aborts makepkg.

when i cd to src/linux-2.6.28/ and run 'make menuconfig' myself, the menuconfig works as expected.

this didn't happen with the 2.6.28 sources, and happens whether or not i apply any patches.

what the hell?!?

p.s.: having posted this, i think perhaps it's better off in the 'PKGBUIL & ABS Support' forum. whoops, sorry.

UPDATE: after mysteriously disappearing, this just as mysteriously reappeared last night. i *think* it's related to using qingy and switching vcs, but... well... weird.

Last edited by kludge (2009-01-24 20:52:31)

kludge · 2009-01-21 09:28:14

also, i just realized that the menuconfig shown when called by makepkg takes up only the upper-left quarter of my console. when i run 'make menuconfig' myself, it takes up the entire screen.

seriously, what the hell?!?

methuselah · 2009-01-21 16:36:05

kludge wrote:

i'm attempting to compile a custom kernel using the PKGBUILD and instructions from Custom Kernel Compilation with ABS. i'm using 2.6.28.1; in fact, all i want to do is enable ACPI debugging messages. after wrangling a deprecated debugging call, i'm already to roll.
except that when makepkg calls 'make menuconfig' inside the build() function, the menu is unuseable. the arrow keys produce '[[B[[C[[A', etc., and i can't select any of the options. i can only get out of the menuconfig with <ctrl>-C, which aborts makepkg.
when i cd to src/linux-2.6.28/ and run 'make menuconfig' myself, the menuconfig works as expected.
this didn't happen with the 2.6.28 sources, and happens whether or not i apply any patches.
what the hell?!?
p.s.: having posted this, i think perhaps it's better off in the 'PKGBUIL & ABS Support' forum. whoops, sorry.

What I do is use the ABS tree's kernel26 PKGBUILD in: /var/abs/core/kernel26/PKGBUILD

Then change this section:

  # load configuration
  yes "" | make config
  # build!

to this section:

  # load configuration
  make menuconfig
  # build!

Then I get a normal menu.

Last edited by methuselah (2009-01-21 16:37:07)

kludge · 2009-01-21 21:32:44

i will try that if all else fails. as it is, i change 'make menuconfig' to 'make xconfig' and that works fine.

i have two reasons for not using the ABS PKGBUILD:

1) while the change i'm trying to make now is fairly trivial, i would like to get together a good method for installing parallel kernels. if i understand correctly, the ABS PKGBUILD will replace the current kernel.

2) kernel compilation takes somewhere on the order of six hours to complete on this box. it takes at least four hours just to fail, so every new method or config change i want to make is a royal p.i.t.a.

methuselah · 2009-01-22 03:49:31

kludge wrote:

i will try that if all else fails. as it is, i change 'make menuconfig' to 'make xconfig' and that works fine.
i have two reasons for not using the ABS PKGBUILD:
1) while the change i'm trying to make now is fairly trivial, i would like to get together a good method for installing parallel kernels. if i understand correctly, the ABS PKGBUILD will replace the current kernel.
2) kernel compilation takes somewhere on the order of six hours to complete on this box. it takes at least four hours just to fail, so every new method or config change i want to make is a royal p.i.t.a.

This is what I do to build a parallel kernel:

I download the patch with "wget -c" - then edit all parts called -ARCH to -ARCHtest - then change name of patch to make it "-ARCHtest"
I change "kernel26.preset" to "kernel26test.preset" - I edit all "26" parts to "26test" within the file to use the new name.
I change "kernel26.install" to "kernel26test.install" - and change these two parts:

KERNEL_VERSION=2.6.28-ARCH

to:

KERNEL_VERSION=2.6.28-ARCHtest

and:

/sbin/mkinitcpio -p kernel26

to:

/sbin/mkinitcpio -p kernel26test

Then edit the PKGBUILD to use the name kernel26test, vmliuz26test, and System.map26test and "26test" for all old "26" parts...... and "-ARCHtest" for all "-ARCH" parts.

.... just patch the same way with the new name "-ARCHtest" then add that "make menuconfig" part.

I also change this part at the top:

_patchname="patch-${pkgver}-${pkgrel}-ARCH"

to

_patchset="-ARCHtestAMD2"

and lastly I change this line in the PKGBUILD:

_kernver="$(make kernelrelease)"

to:

_kernver="${_basekernel}${_patchset}"

Plus, when you use make menuconfig, I use my (-ARCHtest) for my Local Version in General Setup.

I also ditch the md5sums.

..... it seems like a lot at first, but since I have my 3 files already as templates.... all I have to do is edit the patch with "ctrl + h" and rename the patch. (add add my new kernel line in for /boot/grub/menu.lst)

http://i480.photobucket.com/albums/rr16 … t-22-2.png

Last edited by methuselah (2009-01-22 04:08:14)

moz · 2009-01-22 08:50:09

kludge wrote:

i will try that if all else fails. as it is, i change 'make menuconfig' to 'make xconfig' and that works fine.
i have two reasons for not using the ABS PKGBUILD:
1) while the change i'm trying to make now is fairly trivial, i would like to get together a good method for installing parallel kernels. if i understand correctly, the ABS PKGBUILD will replace the current kernel.
2) kernel compilation takes somewhere on the order of six hours to complete on this box. it takes at least four hours just to fail, so every new method or config change i want to make is a royal p.i.t.a.

It takes that long because you're compiling modules and drivers you dont need (if you're compiling just for your own computer). What i do is check all the modules needed by my hardware with lsusb and then search and enable each of them in the kernel configuration menu. But you ought to be careful, disabling the wrong thing might make the kernel unbootable.
After i removed all the drivers i didnt need my compile time went from 5 hrs to 40 mins. Also make sure you enable ccache, heres a guide http://wiki.archlinux.org/index.php/Fas … ache_HOWTO
With ccache i can compile a kernel in aprox 10 mins. And all of this on a 2.6Ghz celeron processor.

Last edited by moz (2009-01-22 08:50:40)

kludge · 2009-01-22 09:02:33

@methuselah and moz: those are great tips. thanks to both of you for them.

however, the point of this thread is to figure why the hell makepkg breaks 'make menuconfig'. anyone?

kludge · 2009-01-23 21:13:52

UPDATE: i was going to file a bug-report about this, but the problem has apparently disappeared of its own accord. ah, what would linux be without the occasional unfathomable mystery?

UPDATE-2: this has cropped up for others, and recurred for me, so i've opened a bug report at http://bugs.archlinux.org/task/13214

Last edited by kludge (2009-02-11 20:21:52)

ajames · 2009-02-14 23:11:29

I registered just to say this because it drove me crazy for a bit just now.

Seems like it was colormake causing the arrow key problems in menuconfig. I was having the exact same problem, then I tried using the full path to make and it worked. I had make aliased to colormake.

kludge · 2009-02-15 02:49:27

@ajames: that sounds like a different issue. check out the bug report linked to in my previous post; i've tracked my issue down to makepkg's use of 'tee' for logging (with the -L flag) conflicting with ncurses.

sounds like you ought to open a new bug!

ajames · 2009-02-15 23:12:40

I read your bug report and I'm not even using makepkg. Still this is something that happens to menuconfig via various means. I've seen posts from 2003 talking about this and apparently never solving it.

The parent process has redirected stdin, that's common to both of our situations. In your case, tee did it; in mine, perl did. Try running it again and then digging around in /proc.

kludge · 2009-02-16 04:42:13

good point. the bug report is specific is 'make menuconfig', but the bug will affect any run of makepkg where you have -L and redirection. (think i got that right.) want to tack a comment onto my bug?

ajames · 2009-02-16 17:33:06

I'm not actually an arch user (mostly debian) so I won't comment on your bug. Feel free to quote me though. I think it's probably that menuconfig isn't getting actual stdin. You can check this by looking at /proc/{pid of mconf}/fd/0. If it links to a tty or something in /dev/pts, then I'm wrong and you've got a different bug. Otherwise, it links to some pipe and you need to find a way to get raw stdin to that mconf process.

I've been thinking about ttys a lot lately. Slim, the login manager, has a somewhat similar bug:
http://developer.berlios.de/bugs/?func= … up_id=2663

Last edited by ajames (2009-02-16 17:34:08)

Arch Linux

#1 2009-01-21 09:13:23

makepkg breaks 'make menuconfig' with 2.6.28.1

#2 2009-01-21 09:28:14

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#3 2009-01-21 16:36:05

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#4 2009-01-21 21:32:44

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#5 2009-01-22 03:49:31

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#6 2009-01-22 08:50:09

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#7 2009-01-22 09:02:33

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#8 2009-01-23 21:13:52

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#9 2009-02-14 23:11:29

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#10 2009-02-15 02:49:27

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#11 2009-02-15 23:12:40

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#12 2009-02-16 04:42:13

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

#13 2009-02-16 17:33:06

Re: makepkg breaks 'make menuconfig' with 2.6.28.1

Board footer