Re: [gentoo-dev] Portage config auto-update - gentoo-dev

From:	Joseph Carter <knghtbrd@×××.org>
To:	Matt Tucker <tuck@×××××××××××××.net>
Cc:	gentoo-dev@g.o
Subject:	Re: [gentoo-dev] Portage config auto-update
Date:	Fri, 07 Feb 2003 10:31:49
Message-Id:	`20030207102608.GC9231@galen.bluecherry.net`
In Reply to:	[gentoo-dev] Portage config auto-update by Matt Tucker

1

On Thu, Feb 06, 2003 at 07:53:53PM -0800, Matt Tucker wrote:

2

[..]

3

> Briefly, the patch implements the following features:

4

>

5

> 	* Alter config file handling (keeping current versions).

6

> 	* Auto-migrate ._cfg files to new structure during Portage upgrade.

7

> 	* Auto-replace unchanged config files when new versions are

8

> 	  installed.

9

> 	* New FEATURE config_automerge tells Portage to automatically merge

10

> 	  config file changes into the current file. Finding a conflict

11

> 	  aborts this merge.

12

> 	* Implement etc-update in python.

13

> 	* Fully implement 'menu' mode in etc-update (using dialog).

14

>       Requires pythondialog

15

>       <http://bugs.gentoo.org/show_bug.cgi?id=15238>.

16

> 	* Support 3-way diffs/merges in etc-update.

17

18

Very cool.  This is almost enough to actually use CVS for files in /etc,

19

or at least RCS.  The reason why RCS might be needed instead of CVS is

20

illustrated using file:

21

22

galen:/etc# find \! -type d | xargs file | egrep -v

23

'text|empty|source|symbolic link'

24

./ld.so.cache:                                  data

25

./ssh/ssh_host_key:                             data

26

./mail/aliases.db:                              Berkeley DB (Hash, version

27

7, native byte-order)

28

./ioctl.save:                                   data

29

./X11/app-defaults/XF86Cfg:                     Bennet Yee's "face" format

30

./X11/rstart/rstartd.real:                      ELF 32-bit LSB executable,

31

Intel 80386, version 1 (SYSV), for GNU/Linux 2.4.0, dynamically linked

32

(uses shared libs), stripped

33

./X11/xdm/chooser:                              ELF 32-bit LSB executable,

34

Intel 80386, version 1 (SYSV), for GNU/Linux 2.4.0, dynamically linked

35

(uses shared libs), stripped

36

./cups/ppd/hp1220c.ppd:                         PPD file, ve

37

./cups/ppds.dat:                                data

38

39

Note that CVS is not fond of symlinks, but tollerates them as long as they

40

are not in the CVS database.  Suggest two trivial functions to manage

41

this: one which prunes nonexistant files from .cvsignore, and one which

42

ensures that any symlinks get listed.

43

44

CVS does not gracefully cope with binary data.  Obviously certain files

45

here are generated automatically and would also get .cvsignore'd.  ELF

46

binaries have absolutely zero place in /etc, according to the FHS,

47

however.

48

49

RCS would allow config file management without dealing with these issues

50

since you'd only maintain RCS files for those things you actually care

51

about.  I would consider this more a bandaid though, I would prefer that

52

etc-update have a config option to maintain .cvsignore files for you so

53

that your CONFIG_PROTECT directories may trivially be stored in CVS, using

54

a helper tool which a sysadmin can use themselves if they like.  In this

55

way, etc-update would merge changes into a temp file, let you edit it, and

56

then overwrite your existing file as an atomic operation.  When you're

57

done, check the results into CVS for safe-keeping.

58

59

60

CVS vendor tags are also an option, however they are more complex to work

61

with in my experience and fewer people will be able to do it well.  Doing

62

the merge into the working copy in the CONFIG_PROTECT'd directory is

63

probably easier for people.  If the files are also checked in with a

64

vendor tag, that'll just make it easier for tools an sysadmins which grok

65

more advanced CVS usage.

66

67

68

I haven't yet looked over your patch extensively (and do not trust myself

69

to not screw something up at this hour), but I figured I'd offer my

70

comments anyway.  Take them with a Syberian salt mine if they seem to be

71

missing something obvious.  =)

72

73

74

[..]

75

> One other thing I wanted to point out is that I renamed portage.py to

76

> portage/__init__.py and added the new functionality in

77

> portage/config_protect.py. This change will allow the portage module to

78

> eventually be broken up into separate smaller modules. I think this

79

> would be a good thing from a maintenance standpoint.

80

81

A good plan, I think, looking at the file in question.

82

83

--

84

Joseph Carter <knghtbrd@×××.org>                 I swallowed your goldfish

85

86

<Overfiend> Don't come crying to me about your "30 minute compiles"!!  I

87

            have to build X uphill both ways!  In the snow!  With bare

88

            feet! And we didn't have compilers!  We had to translate the

89

            C code to mnemonics OURSELVES!

90

<Overfiend> And I was 18 before we even had assemblers!

Gentoo Archives: gentoo-dev

Replies

1	On Thu, Feb 06, 2003 at 07:53:53PM -0800, Matt Tucker wrote:
2	[..]
3	> Briefly, the patch implements the following features:
4	>
5	> * Alter config file handling (keeping current versions).
6	> * Auto-migrate ._cfg files to new structure during Portage upgrade.
7	> * Auto-replace unchanged config files when new versions are
8	> installed.
9	> * New FEATURE config_automerge tells Portage to automatically merge
10	> config file changes into the current file. Finding a conflict
11	> aborts this merge.
12	> * Implement etc-update in python.
13	> * Fully implement 'menu' mode in etc-update (using dialog).
14	> Requires pythondialog
15	> <http://bugs.gentoo.org/show_bug.cgi?id=15238>.
16	> * Support 3-way diffs/merges in etc-update.
17
18	Very cool. This is almost enough to actually use CVS for files in /etc,
19	or at least RCS. The reason why RCS might be needed instead of CVS is
20	illustrated using file:
21
22	galen:/etc# find \! -type d \| xargs file \| egrep -v
23	'text\|empty\|source\|symbolic link'
24	./ld.so.cache: data
25	./ssh/ssh_host_key: data
26	./mail/aliases.db: Berkeley DB (Hash, version
27	7, native byte-order)
28	./ioctl.save: data
29	./X11/app-defaults/XF86Cfg: Bennet Yee's "face" format
30	./X11/rstart/rstartd.real: ELF 32-bit LSB executable,
31	Intel 80386, version 1 (SYSV), for GNU/Linux 2.4.0, dynamically linked
32	(uses shared libs), stripped
33	./X11/xdm/chooser: ELF 32-bit LSB executable,
34	Intel 80386, version 1 (SYSV), for GNU/Linux 2.4.0, dynamically linked
35	(uses shared libs), stripped
36	./cups/ppd/hp1220c.ppd: PPD file, ve
37	./cups/ppds.dat: data
38
39	Note that CVS is not fond of symlinks, but tollerates them as long as they
40	are not in the CVS database. Suggest two trivial functions to manage
41	this: one which prunes nonexistant files from .cvsignore, and one which
42	ensures that any symlinks get listed.
43
44	CVS does not gracefully cope with binary data. Obviously certain files
45	here are generated automatically and would also get .cvsignore'd. ELF
46	binaries have absolutely zero place in /etc, according to the FHS,
47	however.
48
49	RCS would allow config file management without dealing with these issues
50	since you'd only maintain RCS files for those things you actually care
51	about. I would consider this more a bandaid though, I would prefer that
52	etc-update have a config option to maintain .cvsignore files for you so
53	that your CONFIG_PROTECT directories may trivially be stored in CVS, using
54	a helper tool which a sysadmin can use themselves if they like. In this
55	way, etc-update would merge changes into a temp file, let you edit it, and
56	then overwrite your existing file as an atomic operation. When you're
57	done, check the results into CVS for safe-keeping.
58
59
60	CVS vendor tags are also an option, however they are more complex to work
61	with in my experience and fewer people will be able to do it well. Doing
62	the merge into the working copy in the CONFIG_PROTECT'd directory is
63	probably easier for people. If the files are also checked in with a
64	vendor tag, that'll just make it easier for tools an sysadmins which grok
65	more advanced CVS usage.
66
67
68	I haven't yet looked over your patch extensively (and do not trust myself
69	to not screw something up at this hour), but I figured I'd offer my
70	comments anyway. Take them with a Syberian salt mine if they seem to be
71	missing something obvious. =)
72
73
74	[..]
75	> One other thing I wanted to point out is that I renamed portage.py to
76	> portage/__init__.py and added the new functionality in
77	> portage/config_protect.py. This change will allow the portage module to
78	> eventually be broken up into separate smaller modules. I think this
79	> would be a good thing from a maintenance standpoint.
80
81	A good plan, I think, looking at the file in question.
82
83	--
84	Joseph Carter <knghtbrd@×××.org> I swallowed your goldfish
85
86	<Overfiend> Don't come crying to me about your "30 minute compiles"!! I
87	have to build X uphill both ways! In the snow! With bare
88	feet! And we didn't have compilers! We had to translate the
89	C code to mnemonics OURSELVES!
90	<Overfiend> And I was 18 before we even had assemblers!