hgbook
view en/ch04-concepts.xml @ 624:3c5e1c03cc3e
Add XSL stylesheets for DocBook
| author | Dongsheng Song <dongsheng.song@gmail.com> |
|---|---|
| date | Thu Mar 12 15:47:15 2009 +0800 (2009-03-12) |
| parents | b90b024729f1 |
| children | cfdb601a3c8b |
line source
1 <!-- vim: set filetype=docbkxml shiftwidth=2 autoindent expandtab tw=77 : -->
3 <chapter id="chap:concepts">
4 <?dbhtml filename="behind-the-scenes.html"?>
5 <title>Behind the scenes</title>
7 <para>Unlike many revision control systems, the concepts upon which
8 Mercurial is built are simple enough that it's easy to understand
9 how the software really works. Knowing this certainly isn't
10 necessary, but I find it useful to have a <quote>mental
11 model</quote> of what's going on.</para>
13 <para>This understanding gives me confidence that Mercurial has been
14 carefully designed to be both <emphasis>safe</emphasis> and
15 <emphasis>efficient</emphasis>. And just as importantly, if it's
16 easy for me to retain a good idea of what the software is doing
17 when I perform a revision control task, I'm less likely to be
18 surprised by its behaviour.</para>
20 <para>In this chapter, we'll initially cover the core concepts
21 behind Mercurial's design, then continue to discuss some of the
22 interesting details of its implementation.</para>
24 <sect1>
25 <title>Mercurial's historical record</title>
27 <sect2>
28 <title>Tracking the history of a single file</title>
30 <para>When Mercurial tracks modifications to a file, it stores
31 the history of that file in a metadata object called a
32 <emphasis>filelog</emphasis>. Each entry in the filelog
33 contains enough information to reconstruct one revision of the
34 file that is being tracked. Filelogs are stored as files in
35 the <filename role="special"
36 class="directory">.hg/store/data</filename> directory. A
37 filelog contains two kinds of information: revision data, and
38 an index to help Mercurial to find a revision
39 efficiently.</para>
41 <para>A file that is large, or has a lot of history, has its
42 filelog stored in separate data
43 (<quote><literal>.d</literal></quote> suffix) and index
44 (<quote><literal>.i</literal></quote> suffix) files. For
45 small files without much history, the revision data and index
46 are combined in a single <quote><literal>.i</literal></quote>
47 file. The correspondence between a file in the working
48 directory and the filelog that tracks its history in the
49 repository is illustrated in figure <xref
50 linkend="fig:concepts:filelog"/>.</para>
52 <informalfigure id="fig:concepts:filelog">
53 <mediaobject><imageobject><imagedata
54 fileref="filelog"/></imageobject><textobject><phrase>XXX
55 add text</phrase></textobject>
56 <caption><para>Relationships between files in working
57 directory and filelogs in
58 repository</para></caption></mediaobject>
59 </informalfigure>
61 </sect2>
62 <sect2>
63 <title>Managing tracked files</title>
65 <para>Mercurial uses a structure called a
66 <emphasis>manifest</emphasis> to collect together information
67 about the files that it tracks. Each entry in the manifest
68 contains information about the files present in a single
69 changeset. An entry records which files are present in the
70 changeset, the revision of each file, and a few other pieces
71 of file metadata.</para>
73 </sect2>
74 <sect2>
75 <title>Recording changeset information</title>
77 <para>The <emphasis>changelog</emphasis> contains information
78 about each changeset. Each revision records who committed a
79 change, the changeset comment, other pieces of
80 changeset-related information, and the revision of the
81 manifest to use.</para>
83 </sect2>
84 <sect2>
85 <title>Relationships between revisions</title>
87 <para>Within a changelog, a manifest, or a filelog, each
88 revision stores a pointer to its immediate parent (or to its
89 two parents, if it's a merge revision). As I mentioned above,
90 there are also relationships between revisions
91 <emphasis>across</emphasis> these structures, and they are
92 hierarchical in nature.</para>
94 <para>For every changeset in a repository, there is exactly one
95 revision stored in the changelog. Each revision of the
96 changelog contains a pointer to a single revision of the
97 manifest. A revision of the manifest stores a pointer to a
98 single revision of each filelog tracked when that changeset
99 was created. These relationships are illustrated in figure
100 <xref linkend="fig:concepts:metadata"/>.</para>
102 <informalfigure id="fig:concepts:metadata">
103 <mediaobject><imageobject><imagedata
104 fileref="metadata"/></imageobject><textobject><phrase>XXX
105 add text</phrase></textobject><caption><para>Metadata
106 relationships</para></caption>
107 </mediaobject>
108 </informalfigure>
110 <para>As the illustration shows, there is
111 <emphasis>not</emphasis> a <quote>one to one</quote>
112 relationship between revisions in the changelog, manifest, or
113 filelog. If the manifest hasn't changed between two
114 changesets, the changelog entries for those changesets will
115 point to the same revision of the manifest. If a file that
116 Mercurial tracks hasn't changed between two changesets, the
117 entry for that file in the two revisions of the manifest will
118 point to the same revision of its filelog.</para>
120 </sect2>
121 </sect1>
122 <sect1>
123 <title>Safe, efficient storage</title>
125 <para>The underpinnings of changelogs, manifests, and filelogs are
126 provided by a single structure called the
127 <emphasis>revlog</emphasis>.</para>
129 <sect2>
130 <title>Efficient storage</title>
132 <para>The revlog provides efficient storage of revisions using a
133 <emphasis>delta</emphasis> mechanism. Instead of storing a
134 complete copy of a file for each revision, it stores the
135 changes needed to transform an older revision into the new
136 revision. For many kinds of file data, these deltas are
137 typically a fraction of a percent of the size of a full copy
138 of a file.</para>
140 <para>Some obsolete revision control systems can only work with
141 deltas of text files. They must either store binary files as
142 complete snapshots or encoded into a text representation, both
143 of which are wasteful approaches. Mercurial can efficiently
144 handle deltas of files with arbitrary binary contents; it
145 doesn't need to treat text as special.</para>
147 </sect2>
148 <sect2 id="sec:concepts:txn">
149 <title>Safe operation</title>
151 <para>Mercurial only ever <emphasis>appends</emphasis> data to
152 the end of a revlog file. It never modifies a section of a
153 file after it has written it. This is both more robust and
154 efficient than schemes that need to modify or rewrite
155 data.</para>
157 <para>In addition, Mercurial treats every write as part of a
158 <emphasis>transaction</emphasis> that can span a number of
159 files. A transaction is <emphasis>atomic</emphasis>: either
160 the entire transaction succeeds and its effects are all
161 visible to readers in one go, or the whole thing is undone.
162 This guarantee of atomicity means that if you're running two
163 copies of Mercurial, where one is reading data and one is
164 writing it, the reader will never see a partially written
165 result that might confuse it.</para>
167 <para>The fact that Mercurial only appends to files makes it
168 easier to provide this transactional guarantee. The easier it
169 is to do stuff like this, the more confident you should be
170 that it's done correctly.</para>
172 </sect2>
173 <sect2>
174 <title>Fast retrieval</title>
176 <para>Mercurial cleverly avoids a pitfall common to all earlier
177 revision control systems: the problem of <emphasis>inefficient
178 retrieval</emphasis>. Most revision control systems store
179 the contents of a revision as an incremental series of
180 modifications against a <quote>snapshot</quote>. To
181 reconstruct a specific revision, you must first read the
182 snapshot, and then every one of the revisions between the
183 snapshot and your target revision. The more history that a
184 file accumulates, the more revisions you must read, hence the
185 longer it takes to reconstruct a particular revision.</para>
187 <informalfigure id="fig:concepts:snapshot">
188 <mediaobject><imageobject><imagedata
189 fileref="snapshot"/></imageobject><textobject><phrase>XXX
190 add text</phrase></textobject><caption><para>Snapshot of
191 a revlog, with incremental
192 deltas</para></caption></mediaobject>
193 </informalfigure>
195 <para>The innovation that Mercurial applies to this problem is
196 simple but effective. Once the cumulative amount of delta
197 information stored since the last snapshot exceeds a fixed
198 threshold, it stores a new snapshot (compressed, of course),
199 instead of another delta. This makes it possible to
200 reconstruct <emphasis>any</emphasis> revision of a file
201 quickly. This approach works so well that it has since been
202 copied by several other revision control systems.</para>
204 <para>Figure <xref linkend="fig:concepts:snapshot"/> illustrates
205 the idea. In an entry in a revlog's index file, Mercurial
206 stores the range of entries from the data file that it must
207 read to reconstruct a particular revision.</para>
209 <sect3>
210 <title>Aside: the influence of video compression</title>
212 <para>If you're familiar with video compression or have ever
213 watched a TV feed through a digital cable or satellite
214 service, you may know that most video compression schemes
215 store each frame of video as a delta against its predecessor
216 frame. In addition, these schemes use <quote>lossy</quote>
217 compression techniques to increase the compression ratio, so
218 visual errors accumulate over the course of a number of
219 inter-frame deltas.</para>
221 <para>Because it's possible for a video stream to <quote>drop
222 out</quote> occasionally due to signal glitches, and to
223 limit the accumulation of artefacts introduced by the lossy
224 compression process, video encoders periodically insert a
225 complete frame (called a <quote>key frame</quote>) into the
226 video stream; the next delta is generated against that
227 frame. This means that if the video signal gets
228 interrupted, it will resume once the next key frame is
229 received. Also, the accumulation of encoding errors
230 restarts anew with each key frame.</para>
232 </sect3>
233 </sect2>
234 <sect2>
235 <title>Identification and strong integrity</title>
237 <para>Along with delta or snapshot information, a revlog entry
238 contains a cryptographic hash of the data that it represents.
239 This makes it difficult to forge the contents of a revision,
240 and easy to detect accidental corruption.</para>
242 <para>Hashes provide more than a mere check against corruption;
243 they are used as the identifiers for revisions. The changeset
244 identification hashes that you see as an end user are from
245 revisions of the changelog. Although filelogs and the
246 manifest also use hashes, Mercurial only uses these behind the
247 scenes.</para>
249 <para>Mercurial verifies that hashes are correct when it
250 retrieves file revisions and when it pulls changes from
251 another repository. If it encounters an integrity problem, it
252 will complain and stop whatever it's doing.</para>
254 <para>In addition to the effect it has on retrieval efficiency,
255 Mercurial's use of periodic snapshots makes it more robust
256 against partial data corruption. If a revlog becomes partly
257 corrupted due to a hardware error or system bug, it's often
258 possible to reconstruct some or most revisions from the
259 uncorrupted sections of the revlog, both before and after the
260 corrupted section. This would not be possible with a
261 delta-only storage model.</para>
263 </sect2>
264 </sect1>
265 <sect1>
266 <title>Revision history, branching, and merging</title>
268 <para>Every entry in a Mercurial revlog knows the identity of its
269 immediate ancestor revision, usually referred to as its
270 <emphasis>parent</emphasis>. In fact, a revision contains room
271 for not one parent, but two. Mercurial uses a special hash,
272 called the <quote>null ID</quote>, to represent the idea
273 <quote>there is no parent here</quote>. This hash is simply a
274 string of zeroes.</para>
276 <para>In figure <xref linkend="fig:concepts:revlog"/>, you can see
277 an example of the conceptual structure of a revlog. Filelogs,
278 manifests, and changelogs all have this same structure; they
279 differ only in the kind of data stored in each delta or
280 snapshot.</para>
282 <para>The first revision in a revlog (at the bottom of the image)
283 has the null ID in both of its parent slots. For a
284 <quote>normal</quote> revision, its first parent slot contains
285 the ID of its parent revision, and its second contains the null
286 ID, indicating that the revision has only one real parent. Any
287 two revisions that have the same parent ID are branches. A
288 revision that represents a merge between branches has two normal
289 revision IDs in its parent slots.</para>
291 <informalfigure id="fig:concepts:revlog">
292 <mediaobject><imageobject><imagedata
293 fileref="revlog"/></imageobject><textobject><phrase>XXX
294 add text</phrase></textobject></mediaobject>
295 </informalfigure>
297 </sect1>
298 <sect1>
299 <title>The working directory</title>
301 <para>In the working directory, Mercurial stores a snapshot of the
302 files from the repository as of a particular changeset.</para>
304 <para>The working directory <quote>knows</quote> which changeset
305 it contains. When you update the working directory to contain a
306 particular changeset, Mercurial looks up the appropriate
307 revision of the manifest to find out which files it was tracking
308 at the time that changeset was committed, and which revision of
309 each file was then current. It then recreates a copy of each of
310 those files, with the same contents it had when the changeset
311 was committed.</para>
313 <para>The <emphasis>dirstate</emphasis> contains Mercurial's
314 knowledge of the working directory. This details which
315 changeset the working directory is updated to, and all of the
316 files that Mercurial is tracking in the working
317 directory.</para>
319 <para>Just as a revision of a revlog has room for two parents, so
320 that it can represent either a normal revision (with one parent)
321 or a merge of two earlier revisions, the dirstate has slots for
322 two parents. When you use the <command role="hg-cmd">hg
323 update</command> command, the changeset that you update to is
324 stored in the <quote>first parent</quote> slot, and the null ID
325 in the second. When you <command role="hg-cmd">hg
326 merge</command> with another changeset, the first parent
327 remains unchanged, and the second parent is filled in with the
328 changeset you're merging with. The <command role="hg-cmd">hg
329 parents</command> command tells you what the parents of the
330 dirstate are.</para>
332 <sect2>
333 <title>What happens when you commit</title>
335 <para>The dirstate stores parent information for more than just
336 book-keeping purposes. Mercurial uses the parents of the
337 dirstate as <emphasis>the parents of a new
338 changeset</emphasis> when you perform a commit.</para>
340 <informalfigure id="fig:concepts:wdir">
341 <mediaobject><imageobject><imagedata
342 fileref="wdir"/></imageobject><textobject><phrase>XXX
343 add text</phrase></textobject><caption><para>The working
344 directory can have two
345 parents</para></caption></mediaobject>
346 </informalfigure>
348 <para>Figure <xref linkend="fig:concepts:wdir"/> shows the
349 normal state of the working directory, where it has a single
350 changeset as parent. That changeset is the
351 <emphasis>tip</emphasis>, the newest changeset in the
352 repository that has no children.</para>
354 <informalfigure id="fig:concepts:wdir-after-commit">
355 <mediaobject><imageobject><imagedata
356 fileref="wdir-after-commit"/></imageobject><textobject><phrase>XXX
357 add text</phrase></textobject><caption><para>The working
358 directory gains new parents after a
359 commit</para></caption></mediaobject>
360 </informalfigure>
362 <para>It's useful to think of the working directory as
363 <quote>the changeset I'm about to commit</quote>. Any files
364 that you tell Mercurial that you've added, removed, renamed,
365 or copied will be reflected in that changeset, as will
366 modifications to any files that Mercurial is already tracking;
367 the new changeset will have the parents of the working
368 directory as its parents.</para>
370 <para>After a commit, Mercurial will update the parents of the
371 working directory, so that the first parent is the ID of the
372 new changeset, and the second is the null ID. This is shown
373 in figure <xref linkend="fig:concepts:wdir-after-commit"/>.
374 Mercurial
375 doesn't touch any of the files in the working directory when
376 you commit; it just modifies the dirstate to note its new
377 parents.</para>
379 </sect2>
380 <sect2>
381 <title>Creating a new head</title>
383 <para>It's perfectly normal to update the working directory to a
384 changeset other than the current tip. For example, you might
385 want to know what your project looked like last Tuesday, or
386 you could be looking through changesets to see which one
387 introduced a bug. In cases like this, the natural thing to do
388 is update the working directory to the changeset you're
389 interested in, and then examine the files in the working
390 directory directly to see their contents as they were when you
391 committed that changeset. The effect of this is shown in
392 figure <xref linkend="fig:concepts:wdir-pre-branch"/>.</para>
394 <informalfigure id="fig:concepts:wdir-pre-branch">
395 <mediaobject><imageobject><imagedata
396 fileref="wdir-pre-branch"/></imageobject><textobject><phrase>XXX
397 add text</phrase></textobject><caption><para>The working
398 directory, updated to an older
399 changeset</para></caption></mediaobject>
400 </informalfigure>
402 <para>Having updated the working directory to an older
403 changeset, what happens if you make some changes, and then
404 commit? Mercurial behaves in the same way as I outlined
405 above. The parents of the working directory become the
406 parents of the new changeset. This new changeset has no
407 children, so it becomes the new tip. And the repository now
408 contains two changesets that have no children; we call these
409 <emphasis>heads</emphasis>. You can see the structure that
410 this creates in figure <xref
411 linkend="fig:concepts:wdir-branch"/>.</para>
413 <informalfigure id="fig:concepts:wdir-branch">
414 <mediaobject><imageobject><imagedata
415 fileref="wdir-branch"/></imageobject><textobject><phrase>XXX
416 add text</phrase></textobject><caption><para>After a
417 commit made while synced to an older
418 changeset</para></caption></mediaobject>
419 </informalfigure>
421 <note>
422 <para> If you're new to Mercurial, you should keep in mind a
423 common <quote>error</quote>, which is to use the <command
424 role="hg-cmd">hg pull</command> command without any
425 options. By default, the <command role="hg-cmd">hg
426 pull</command> command <emphasis>does not</emphasis>
427 update the working directory, so you'll bring new changesets
428 into your repository, but the working directory will stay
429 synced at the same changeset as before the pull. If you
430 make some changes and commit afterwards, you'll thus create
431 a new head, because your working directory isn't synced to
432 whatever the current tip is.</para>
434 <para> I put the word <quote>error</quote> in quotes because
435 all that you need to do to rectify this situation is
436 <command role="hg-cmd">hg merge</command>, then <command
437 role="hg-cmd">hg commit</command>. In other words, this
438 almost never has negative consequences; it just surprises
439 people. I'll discuss other ways to avoid this behaviour,
440 and why Mercurial behaves in this initially surprising way,
441 later on.</para>
442 </note>
444 </sect2>
445 <sect2>
446 <title>Merging heads</title>
448 <para>When you run the <command role="hg-cmd">hg merge</command>
449 command, Mercurial leaves the first parent of the working
450 directory unchanged, and sets the second parent to the
451 changeset you're merging with, as shown in figure <xref
452 linkend="fig:concepts:wdir-merge"/>.</para>
454 <informalfigure id="fig:concepts:wdir-merge">
455 <mediaobject><imageobject><imagedata
456 fileref="wdir-merge"/></imageobject><textobject><phrase>XXX
457 add text</phrase></textobject><caption><para>Merging two
458 heads</para></caption></mediaobject>
459 </informalfigure>
461 <para>Mercurial also has to modify the working directory, to
462 merge the files managed in the two changesets. Simplified a
463 little, the merging process goes like this, for every file in
464 the manifests of both changesets.</para>
465 <itemizedlist>
466 <listitem><para>If neither changeset has modified a file, do
467 nothing with that file.</para>
468 </listitem>
469 <listitem><para>If one changeset has modified a file, and the
470 other hasn't, create the modified copy of the file in the
471 working directory.</para>
472 </listitem>
473 <listitem><para>If one changeset has removed a file, and the
474 other hasn't (or has also deleted it), delete the file
475 from the working directory.</para>
476 </listitem>
477 <listitem><para>If one changeset has removed a file, but the
478 other has modified the file, ask the user what to do: keep
479 the modified file, or remove it?</para>
480 </listitem>
481 <listitem><para>If both changesets have modified a file,
482 invoke an external merge program to choose the new
483 contents for the merged file. This may require input from
484 the user.</para>
485 </listitem>
486 <listitem><para>If one changeset has modified a file, and the
487 other has renamed or copied the file, make sure that the
488 changes follow the new name of the file.</para>
489 </listitem></itemizedlist>
490 <para>There are more details&emdash;merging has plenty of corner
491 cases&emdash;but these are the most common choices that are
492 involved in a merge. As you can see, most cases are
493 completely automatic, and indeed most merges finish
494 automatically, without requiring your input to resolve any
495 conflicts.</para>
497 <para>When you're thinking about what happens when you commit
498 after a merge, once again the working directory is <quote>the
499 changeset I'm about to commit</quote>. After the <command
500 role="hg-cmd">hg merge</command> command completes, the
501 working directory has two parents; these will become the
502 parents of the new changeset.</para>
504 <para>Mercurial lets you perform multiple merges, but you must
505 commit the results of each individual merge as you go. This
506 is necessary because Mercurial only tracks two parents for
507 both revisions and the working directory. While it would be
508 technically possible to merge multiple changesets at once, the
509 prospect of user confusion and making a terrible mess of a
510 merge immediately becomes overwhelming.</para>
512 </sect2>
513 </sect1>
514 <sect1>
515 <title>Other interesting design features</title>
517 <para>In the sections above, I've tried to highlight some of the
518 most important aspects of Mercurial's design, to illustrate that
519 it pays careful attention to reliability and performance.
520 However, the attention to detail doesn't stop there. There are
521 a number of other aspects of Mercurial's construction that I
522 personally find interesting. I'll detail a few of them here,
523 separate from the <quote>big ticket</quote> items above, so that
524 if you're interested, you can gain a better idea of the amount
525 of thinking that goes into a well-designed system.</para>
527 <sect2>
528 <title>Clever compression</title>
530 <para>When appropriate, Mercurial will store both snapshots and
531 deltas in compressed form. It does this by always
532 <emphasis>trying to</emphasis> compress a snapshot or delta,
533 but only storing the compressed version if it's smaller than
534 the uncompressed version.</para>
536 <para>This means that Mercurial does <quote>the right
537 thing</quote> when storing a file whose native form is
538 compressed, such as a <literal>zip</literal> archive or a JPEG
539 image. When these types of files are compressed a second
540 time, the resulting file is usually bigger than the
541 once-compressed form, and so Mercurial will store the plain
542 <literal>zip</literal> or JPEG.</para>
544 <para>Deltas between revisions of a compressed file are usually
545 larger than snapshots of the file, and Mercurial again does
546 <quote>the right thing</quote> in these cases. It finds that
547 such a delta exceeds the threshold at which it should store a
548 complete snapshot of the file, so it stores the snapshot,
549 again saving space compared to a naive delta-only
550 approach.</para>
552 <sect3>
553 <title>Network recompression</title>
555 <para>When storing revisions on disk, Mercurial uses the
556 <quote>deflate</quote> compression algorithm (the same one
557 used by the popular <literal>zip</literal> archive format),
558 which balances good speed with a respectable compression
559 ratio. However, when transmitting revision data over a
560 network connection, Mercurial uncompresses the compressed
561 revision data.</para>
563 <para>If the connection is over HTTP, Mercurial recompresses
564 the entire stream of data using a compression algorithm that
565 gives a better compression ratio (the Burrows-Wheeler
566 algorithm from the widely used <literal>bzip2</literal>
567 compression package). This combination of algorithm and
568 compression of the entire stream (instead of a revision at a
569 time) substantially reduces the number of bytes to be
570 transferred, yielding better network performance over almost
571 all kinds of network.</para>
573 <para>(If the connection is over <command>ssh</command>,
574 Mercurial <emphasis>doesn't</emphasis> recompress the
575 stream, because <command>ssh</command> can already do this
576 itself.)</para>
578 </sect3>
579 </sect2>
580 <sect2>
581 <title>Read/write ordering and atomicity</title>
583 <para>Appending to files isn't the whole story when it comes to
584 guaranteeing that a reader won't see a partial write. If you
585 recall figure <xref linkend="fig:concepts:metadata"/>,
586 revisions in the
587 changelog point to revisions in the manifest, and revisions in
588 the manifest point to revisions in filelogs. This hierarchy
589 is deliberate.</para>
591 <para>A writer starts a transaction by writing filelog and
592 manifest data, and doesn't write any changelog data until
593 those are finished. A reader starts by reading changelog
594 data, then manifest data, followed by filelog data.</para>
596 <para>Since the writer has always finished writing filelog and
597 manifest data before it writes to the changelog, a reader will
598 never read a pointer to a partially written manifest revision
599 from the changelog, and it will never read a pointer to a
600 partially written filelog revision from the manifest.</para>
602 </sect2>
603 <sect2>
604 <title>Concurrent access</title>
606 <para>The read/write ordering and atomicity guarantees mean that
607 Mercurial never needs to <emphasis>lock</emphasis> a
608 repository when it's reading data, even if the repository is
609 being written to while the read is occurring. This has a big
610 effect on scalability; you can have an arbitrary number of
611 Mercurial processes safely reading data from a repository
612 safely all at once, no matter whether it's being written to or
613 not.</para>
615 <para>The lockless nature of reading means that if you're
616 sharing a repository on a multi-user system, you don't need to
617 grant other local users permission to
618 <emphasis>write</emphasis> to your repository in order for
619 them to be able to clone it or pull changes from it; they only
620 need <emphasis>read</emphasis> permission. (This is
621 <emphasis>not</emphasis> a common feature among revision
622 control systems, so don't take it for granted! Most require
623 readers to be able to lock a repository to access it safely,
624 and this requires write permission on at least one directory,
625 which of course makes for all kinds of nasty and annoying
626 security and administrative problems.)</para>
628 <para>Mercurial uses locks to ensure that only one process can
629 write to a repository at a time (the locking mechanism is safe
630 even over filesystems that are notoriously hostile to locking,
631 such as NFS). If a repository is locked, a writer will wait
632 for a while to retry if the repository becomes unlocked, but
633 if the repository remains locked for too long, the process
634 attempting to write will time out after a while. This means
635 that your daily automated scripts won't get stuck forever and
636 pile up if a system crashes unnoticed, for example. (Yes, the
637 timeout is configurable, from zero to infinity.)</para>
639 <sect3>
640 <title>Safe dirstate access</title>
642 <para>As with revision data, Mercurial doesn't take a lock to
643 read the dirstate file; it does acquire a lock to write it.
644 To avoid the possibility of reading a partially written copy
645 of the dirstate file, Mercurial writes to a file with a
646 unique name in the same directory as the dirstate file, then
647 renames the temporary file atomically to
648 <filename>dirstate</filename>. The file named
649 <filename>dirstate</filename> is thus guaranteed to be
650 complete, not partially written.</para>
652 </sect3>
653 </sect2>
654 <sect2>
655 <title>Avoiding seeks</title>
657 <para>Critical to Mercurial's performance is the avoidance of
658 seeks of the disk head, since any seek is far more expensive
659 than even a comparatively large read operation.</para>
661 <para>This is why, for example, the dirstate is stored in a
662 single file. If there were a dirstate file per directory that
663 Mercurial tracked, the disk would seek once per directory.
664 Instead, Mercurial reads the entire single dirstate file in
665 one step.</para>
667 <para>Mercurial also uses a <quote>copy on write</quote> scheme
668 when cloning a repository on local storage. Instead of
669 copying every revlog file from the old repository into the new
670 repository, it makes a <quote>hard link</quote>, which is a
671 shorthand way to say <quote>these two names point to the same
672 file</quote>. When Mercurial is about to write to one of a
673 revlog's files, it checks to see if the number of names
674 pointing at the file is greater than one. If it is, more than
675 one repository is using the file, so Mercurial makes a new
676 copy of the file that is private to this repository.</para>
678 <para>A few revision control developers have pointed out that
679 this idea of making a complete private copy of a file is not
680 very efficient in its use of storage. While this is true,
681 storage is cheap, and this method gives the highest
682 performance while deferring most book-keeping to the operating
683 system. An alternative scheme would most likely reduce
684 performance and increase the complexity of the software, each
685 of which is much more important to the <quote>feel</quote> of
686 day-to-day use.</para>
688 </sect2>
689 <sect2>
690 <title>Other contents of the dirstate</title>
692 <para>Because Mercurial doesn't force you to tell it when you're
693 modifying a file, it uses the dirstate to store some extra
694 information so it can determine efficiently whether you have
695 modified a file. For each file in the working directory, it
696 stores the time that it last modified the file itself, and the
697 size of the file at that time.</para>
699 <para>When you explicitly <command role="hg-cmd">hg
700 add</command>, <command role="hg-cmd">hg remove</command>,
701 <command role="hg-cmd">hg rename</command> or <command
702 role="hg-cmd">hg copy</command> files, Mercurial updates the
703 dirstate so that it knows what to do with those files when you
704 commit.</para>
706 <para>When Mercurial is checking the states of files in the
707 working directory, it first checks a file's modification time.
708 If that has not changed, the file must not have been modified.
709 If the file's size has changed, the file must have been
710 modified. If the modification time has changed, but the size
711 has not, only then does Mercurial need to read the actual
712 contents of the file to see if they've changed. Storing these
713 few extra pieces of information dramatically reduces the
714 amount of data that Mercurial needs to read, which yields
715 large performance improvements compared to other revision
716 control systems.</para>
718 </sect2>
719 </sect1>
720 </chapter>
722 <!--
723 local variables:
724 sgml-parent-document: ("00book.xml" "book" "chapter")
725 end:
726 -->