automate lfs-commands creation

[markh@…]

change the script.xml files to include a file called script-whatever.xml and then put just the script file in that. We should then be able to simply sed the four or five entities across using a script and so be able to pull the scripts out of cvs with one command.

[gerard@…]

if we were to do this here, we may as well do this with every package installation. I'd like to look for a better method first to extract the commands.

I'm not totally against a dedicated command instructions file, but it is a lot of work and the only reason why we'd do it, is to facilitate the lfs-commands stuff. Even with dedicated instructions files, we'd still have to strip the screen and para tags, so a script has to be created regardless. Then we better just keep it the way it is and deal with the *-inst.xml files directly.

I'm voting to close this

[timothy@…]

I made a script to build lfs-commands automagically:

​http://linuxfromscratch.org/~timothy/lfscmd/lfscmd-0.1.tar.bz2

1.chroot doesn't conform to the <screen><userinput> stuff. I

think that <screen> should go inside the chrootcmd.xml file, not in chroot.xml as it currently is. I'll change this unless anyone has any objections.

2.commands that use &*-version;, such as those found in

binutils-inst.xml, don't evaluate to their version number in lfs-commands.

Possible solutions:

(1) grep the version number in entities, cut it, and replace

it with $*-version on the fly.

(2) Have a list of &-*version;'s and replace them from a

manually-maintained list.

(3) Hardcode the version number in the commands in LFS BOOK,

as some (such as 2 of the findutils patch lines) already are.

3. commands which escape characters, such as '\(' and '\)' in the sed commands of sysvinit, are generally getting fscked up. I have a feeling that BASH is involved. IMO, these commands would look better with '@' as a deliminator, and the need for escaping '(' would dissapear. I'll change this as well, unless anyone has any objections.

4. it is slow. Not as slow as parsing the xml into HTML, and not as slow as "should i get up and do something else?", but slow as in "hmm, i should re-paint my ceiling."

this lack of speed is attributed to the lfscmd-parse script. I had made this script because my lack of awk/sed talent was resulting in 80% correct lfs-commands and 20% whitespace :)

lfscmd-parse would be faster if written in C. I can do that, but i really don't feel like it, i can live with twiddling my thumbs an extra 15 seconds.

Anyhoo, this was a long comment, eh? Tell me what you think.

[markh@…]

1. No, as long as you check that the syntax is still valid, I don't object to putting the <screen> tag inside chroot-cmd.xml

2. If we're automating this, the following is the only solution: (1) grep the version number in entities, cut it, and replace

it with $*-version on the fly.

The option (3) won't fly, indeed I've just fixed the hard-coded values you referred to!

3. That looks good to me. More shell independent as well I think. I thought we were going to do this a while ago, obviously it never got done!

4. Personally I didn't think the script was that slow. Certainly not compared to the length of time it takes openjade to generate the book itself..

I like this a lot. I reckon this'll put us well back on the way to having the lfs-commands tarballs again. Kudos to Timothy!

[timothy@…]

New version:

​http://linuxfromscratch.org/~timothy/lfscmd/lfscmd-0.2.tar.bz2

Added an lfscmd-check script to validate that all entities in lfs-commands are being resolved. lfscmd-check is not used by the other scripts.

1. Done.

2. Done. lfscmd now replaces entities on the fly, as long as they are in entities/ or index.xml. Otherwise, a WARNING is sent to stderr.

3. The other sed commands in the LFS book use '%' as a deliminator. They are modifying directory names, so '%' is a nice alternative to escaping /. I simplified the sed command for Sysvinit, so the problems with '\(' and '\)' are resolved for the time being.

4. It is slower now, because I added a bunch of junk in order to cut the entities on the fly. lfscmd-check will handle some of this load in the next version.

Known Bugs:

1. The commands in Vim use 12 escapes. None of them show up in lfs-commands. I vote that we get rid of Vim :) j/k I'll try to solve the issue with character escapes in the next version.

2. The bottom two lines in fstab, which are not surrounded by <userinput> tags, are ignored. I will make a workaround for this in the next version.

[timothy@…]

New *stable* version:

​http://linuxfromscratch.org/~timothy/lfscmd/lfscmd-0.3.tar.bz2

All known bugs are fixed. lfs-commands render flawlessly.

lfscmd can handle escape characters now. I suggest that lfscmd be added to the nightly maintenance work. Closing bug #177.

[timothy@…]

New version: 0.4

Removed lfscmd-{check,parse} Requires filterin ​http://linuxfromscratch.org/~timothy/filterin Runs in ~6 seconds now.

[timothy@…]

Forgot to give the URL for 0.4: ​http://linuxfromscratch.org/~timothy/lfscmd/lfscmd-0.4.tar.bz2

[timothy@…]

CHANGES:

o Rewritten in C. o Depends on libxml2. o No longer requires filterin. o Works with both the LFS and BLFS books. o Can output to either stdout or files (requested by Archaic).

​http://linuxfromscratch.org/~timothy/lfscmd/lfscmd-0.5.tar.bz2

[timothy@…]

Version increment (1.0): ​http://linuxfromscratch.org/~timothy/lfscmd/lfscmd-1.0.tar.bz2