1
*****************************************
2
Opportunities for improvement on GNU Arch
3
*****************************************
5
[note that this document is rather out of date in 2005-08]
7
GNU Arch is one influence on bazaar-ng. There are several things we
8
would change from Arch in Bazaar to (we hope) improve the user
11
The core design of Arch is good, brilliant even. It can scale from
12
small projects too large ones, and is a good foundation for building
13
tools on top. However, the design is far too complex, both in
14
concepts and execution. So the plan is to cut out as many things as
15
we can, add a few other good concepts from other systems, and try to
16
make it into a whole that is consistent and understandable.
24
No other system is able to express this valuable idea: "I merged all
25
these changes from other people; here is the result."
27
However, it should *also* be possible to bring in perfect-fit
28
patches without creating a new commit.
32
Find a common ancestor on diverged and cross-merged branches.
34
* Apply isolated changesets.
36
We should extend this by having a good way to send changesets by
37
email, preferably readable even by people who are not using Arch.
39
* GPG signing of commits.
41
Open source hackers almost all have GPG keys already, and GPG deals
42
with a lot of PKI functions to do with propagating, signing and
45
Signed commits are interesting in many ways, not least of which in
46
detecting intrusion to code servers.
48
* Anonymous downloads can be done without an active server.
50
Good for security; also very good for people who do not have a
51
permnanently-connected machine on which they can install their own
52
software, or which is very tightly secured.
54
It's neat that you can upload over only sftp/ftp, but I'm not sure
55
it's really worth the hassle; getting properly atomic operations
56
over remote-file protocols is hard.
58
* Clean and transparent storage format.
60
This is a neat hack, and gives people assurance that they can get
61
their data back out again even if the tool disappears. Very nice.
62
(Bazaar-NG won't keep the exact same format, but the ideas will be
65
* Relatively easily parseable/scriptable shell interface. Good for
66
people writing web/emacs/editor/IDE interfaces, or scripts based it.
68
* Automatically build (and hardlink) revision libraries, with
71
I don't know how many people want *every* revision in a library, but
72
it can be handy to have a few key ones.
74
In general making use of hardlinks when they are available and safe
77
* Rely on ssh for remote access, authentication, and confidentiality.
79
* Patch headers separate from patch bodies. (Sometimes you only want
82
* Autogeneration of Changelogs -- but should be in GNU format, at
83
least optionally. I'm not convinced auto-updating them in the tree
84
is worthwhile; it makes merges weird.
88
It seems useful to prevent accidental commits to things that are
89
meant to be stable. However, the set-once nature of sealing is
90
undesirable, because people can make mistakes or want to seal more
93
One possibility is to have a voluntary write-protect flag set on
94
branches that should not normally be updated. One can remove the
95
flag if it turns out it was set wrongly.
97
* ``resolved`` command in Bazaar-1.1
99
Good for preventing accidental breakage.
101
* Multi-level undo -- though could perhaps be more understandable,
102
perhaps through ``undo-history``.
108
One lesson from usability design is that it does not always work to
109
have a complex model and then try to hide complexity in the user
110
interface. If you want something to be a joy to use, that must be
111
designed in from the bottom up.
113
(Some developers may react to tla by thinking "eww, how gross" on
114
particular points. As much as possible we might like to fix these.)
116
* General impression that the tool is telling you how to run your life.
118
* Non-standard terminology
120
Arch uses terms like "version" and "category" in ways that are
121
confusing to people accustomed to other version control systems.
124
Therefore: development proceeds on a *branch*, which is a series of
125
*revisions*. Simple and obvious.
129
* Command-line options are wierdly inconsistent with both other
130
systems, with each others, and with what people would like to do.
131
For example, I would think the obvious usage is ``bzr diff [FILE]``,
132
but ``tla diff`` does not let you specify a file at all.
134
Most commands should take filenames as their argument: log, diff,
137
* Despite having too many commands, there are massive and glaring
138
gaps, such reverting a single file or a tree.
140
* Commands are too different from what people are used to in CVS, and
141
often not for a good reason.
143
* Identifiers are too long. In part this is because Arch tries to
144
have identifiers which are both human-assigned and universally unique.
146
* Archive names are probably unnecessary.
148
* Part of the reason for complexity in archives is that the Arch
149
design wants to be able to go and find patches on other branches at
150
a later time. (This is not really implemented or used at the
153
I think the complexity is unjustified: changesets and revisions have
154
universally unique names so they can simply be archived, either on
155
the machine of the person who wants them or on a central site like
158
* The tool is *unforgiving*; if people create a branch with the wrong
159
name it will be around forever.
161
* Branches are heaviweight; a record always persists in the archive.
162
Sometimes it is good to create micro-branches, try something out,
163
and then discard them. If nobody wants the changes, there is no
164
reason for the tool to keep them.
166
* Working offline requires creating a new branch and merging back and
167
forth. This is both more work than it should be, and also polutes
168
the "story" told by branching.
170
As much as possible, the *accidental* difference of the location of
171
the repository should not effect the *semantics* of branches.
173
(However, some merging may obviously be necessary when there is
176
* Archive registration. This causes confusion and is unnecessary.
178
Proposed solutions such as archive aliases or an additional command
179
to register-and-get make it worse.
181
* Wierd file names (``++`` and ``,,``, which persist in user
182
directories and cause breakage of many tools. Gives a bad
183
impression, and it's even worse when people have to interact with
186
* Overly-long identifiers. (One advantage of pointing to branches
187
using filenames or URLs is that the length of the path depends on
188
how close it is to the users location, and they can more easily use
190
* Too slow by default.
192
Arch can be made fast, but in the hands of a nonexpert user it is
193
often slow. For most users, disk is cheaper than CPU time, which is
194
cheaper than network roundtrips. The performance model should be
195
transparent -- users should not be surprised that something is slow.
197
* Tagging onto branches.
199
Unifying tags and commits is interesting, but the result is hard to
200
mentally model; even Arch maintainers can't say exactly how it is
201
supposed to work in some cases.
203
* Reinventing the world from scratch in libhackerlab/frob/pika/xl.
205
Those are all fine projects and may be useful in the future, but
206
they are totally unnecessary to write a great version control
207
system. It is not an enormous project; it is not CPU-cycle
208
critical; something like Python will be fine.
210
* Lack (for the moment) of an active server.
212
Given that network traffic is the most expensive thing, we can
213
possibly get a better solution by having intelligence on both sides
214
of the link. Suppose we want to get just one file from a previous
217
* Poor Windows/Mac support.
219
Even though many developers only work on Linux, this still holds a
220
tool back. The reason is this: at least some projects have some
221
developers on Windows some of the time. Those projects can't switch
222
to Arch. Most people want to only learn one tool deeply, so it
225
Don't make any overly Unixy assumptions. Avoid too-cute filesystem
228
Being in Python should help with portability: people do need to
229
install it, but many developers will already have it and the total
230
burden is possibly less than that of installing C requisite
233
* Quirky filename support.
235
Files with non-ascii names, or names containing whitespace tend to
236
be handled poorly, perhaps partly because of arch's shell heritage.
238
By swallowing XML we do at least get automatic quoting of wierd
239
strings, and we will always use UTF-8 for internal storage.
241
* Complex file-id-tagging
243
Nobody should be expected to understand this. There are two basic
244
cases: people want to auto-add everything, and want to add by hand.
245
Both can be reasonably accomodated in a simpler system.
247
* Complex naming-convention regexps in ``.arch-inventory`` and
248
``{arch}/id-tagging-method``. (The fact that there are two
249
overlapping mechanisms with very different names is also bad.)
251
All this complexity basically just comes down to versioned, ignored,
252
unknown, the same as in every other system. So we might as well
255
There are relatively few cases where regexps help more than globs,
256
and people do find them more complex. Even experienced users can
257
forget to escape ``\.``. We can have a bit of flexibility with
258
(say) zsh-style extended globs like ``*.(pyo|pyc)``.
260
* Some files inside ``{arch}`` are meant to be edited by the user, and
261
some are not. This is a flaw common to other systems, including
262
Bitkeeper. The user should be clear on whether they should touch
263
things in a directory or not.
265
* Source-librarian function works poorly.
267
It is not the place of a tool to force people to stay organized; it
268
should just facilitate it. In any case, a library without
269
descriptive text is of little use. So bazaar-ng does not force
270
three-level naming but rather lets people arrange their own trees,
271
and put on their own descriptions (either within the tree, or by
272
e.g. having a wiki page listing branches, descriptions and URLs.)
274
* Whining about inode mismatches on pristines/revlibs.
276
It's fine that there is validation, but the tool should not show off
277
its limitations. Just do the right thing.
279
* More generally, not quite enough consistency/safety checking.
281
* Unclear what commands work on subdirs and what works on the whole
284
* Hard to share work on a single branch -- though still not really too
287
* Lack of partial commits of added/deleted files.
289
* Separate id tags for each file; simple implementation but probably
290
costs too much disk space.
292
* Way too many deeply-nested directories; should be just one.
294
* ``.listing`` files are ugly and a point of failure. They can cause
295
trouble on some servers which limit access to dot files.
297
Isn't it possible to have the top-level file be predictable and find
298
everything else needed from there?
300
* Summary separate from log message.
302
Simpler to just have one message, and let people extract the first
303
line/sentence if they wish.
305
Rather than 'keywords', let arbitrary properties be attached to the
306
revision at the time of commit.
310
Simpler disconnected operation
311
------------------------------
313
A basic distributed VCS operation is to make it easy to work on an
314
offline laptop. Arch can do this in a few ways, but none of them are
317
http://wiki.gnuarch.org/moin.cgi/mini_5fTravellingOftenWithArch
319
Yaron Minsky writes (2005-01-18):
321
I was wondering what people considered to be a good setup for using
322
Arch on a laptop. Here's the basic situation. I have a few projects
323
that reside in arch repositories on my desktop computer. Basically,
324
I'd like to be able to do commits from my laptop, and have those
325
commits eventually migrate up to the main repository. I understand
326
that the right way of doing this is to set up archives on the laptop.
327
But what's the cleanest way of doing this? And is there some way of
328
making the commits I do on the laptop show up cleanly and individually
329
on the desktop once they are merged in?
335
baz default is much less strict.
337
Much of tla depends on being able to categorize files. Some hangovers
338
from larch -- eg precious and backup are essentially the same. junk
339
is never deleted today.
341
Automatic version control with 'untagged-source source'. But this is
347
- having the feature at all
348
- complex way to define it
350
Default of 166 lines.
352
Remove id-tagging-method command or at most make it read-only. If
353
people really want to use deprecated methods they can just edit the
356
So we can ship a default id-tagging which works the same as CVS/Svn:
357
give warnings for files that are not known to be junk. This is the
358
default in baz right now.
360
Also we have .arch-inventory, which is per-directory.
364
Why not have 'baz ignore FILENAME'? To remove ignores, perhaps you
365
have to edit the .arch-inventory. Print "FILTER added to
366
PATH/.arch-inventory"; create and baz-add this file if it doesn't.
368
Docs should perhaps emphasize .arch-inventory as the basic method and
369
only mention =tagging-method as an advanced topic.
373
Should this really be regexps, or just file globs?
b'\\ No newline at end of file'
added
removed
doc/todo-from-arch.txt
