hgbook
view en/concepts.tex @ 275:96ea24a916f9
Merge with myself.
author | Bryan O'Sullivan <bos@serpentine.com> |
---|---|
date | Mon Nov 26 20:42:36 2007 -0800 (2007-11-26) |
parents | 3eaa02162e90 |
children | 97e929385442 |
line source
1 \chapter{Behind the scenes}
2 \label{chap:concepts}
4 Unlike many revision control systems, the concepts upon which
5 Mercurial is built are simple enough that it's easy to understand how
6 the software really works. Knowing this certainly isn't necessary,
7 but I find it useful to have a ``mental model'' of what's going on.
9 This understanding gives me confidence that Mercurial has been
10 carefully designed to be both \emph{safe} and \emph{efficient}. And
11 just as importantly, if it's easy for me to retain a good idea of what
12 the software is doing when I perform a revision control task, I'm less
13 likely to be surprised by its behaviour.
15 In this chapter, we'll initially cover the core concepts behind
16 Mercurial's design, then continue to discuss some of the interesting
17 details of its implementation.
19 \section{Mercurial's historical record}
21 \subsection{Tracking the history of a single file}
23 When Mercurial tracks modifications to a file, it stores the history
24 of that file in a metadata object called a \emph{filelog}. Each entry
25 in the filelog contains enough information to reconstruct one revision
26 of the file that is being tracked. Filelogs are stored as files in
27 the \sdirname{.hg/store/data} directory. A filelog contains two kinds
28 of information: revision data, and an index to help Mercurial to find
29 a revision efficiently.
31 A file that is large, or has a lot of history, has its filelog stored
32 in separate data (``\texttt{.d}'' suffix) and index (``\texttt{.i}''
33 suffix) files. For small files without much history, the revision
34 data and index are combined in a single ``\texttt{.i}'' file. The
35 correspondence between a file in the working directory and the filelog
36 that tracks its history in the repository is illustrated in
37 figure~\ref{fig:concepts:filelog}.
39 \begin{figure}[ht]
40 \centering
41 \grafix{filelog}
42 \caption{Relationships between files in working directory and
43 filelogs in repository}
44 \label{fig:concepts:filelog}
45 \end{figure}
47 \subsection{Managing tracked files}
49 Mercurial uses a structure called a \emph{manifest} to collect
50 together information about the files that it tracks. Each entry in
51 the manifest contains information about the files present in a single
52 changeset. An entry records which files are present in the changeset,
53 the revision of each file, and a few other pieces of file metadata.
55 \subsection{Recording changeset information}
57 The \emph{changelog} contains information about each changeset. Each
58 revision records who committed a change, the changeset comment, other
59 pieces of changeset-related information, and the revision of the
60 manifest to use.
62 \subsection{Relationships between revisions}
64 Within a changelog, a manifest, or a filelog, each revision stores a
65 pointer to its immediate parent (or to its two parents, if it's a
66 merge revision). As I mentioned above, there are also relationships
67 between revisions \emph{across} these structures, and they are
68 hierarchical in nature.
70 For every changeset in a repository, there is exactly one revision
71 stored in the changelog. Each revision of the changelog contains a
72 pointer to a single revision of the manifest. A revision of the
73 manifest stores a pointer to a single revision of each filelog tracked
74 when that changeset was created. These relationships are illustrated
75 in figure~\ref{fig:concepts:metadata}.
77 \begin{figure}[ht]
78 \centering
79 \grafix{metadata}
80 \caption{Metadata relationships}
81 \label{fig:concepts:metadata}
82 \end{figure}
84 As the illustration shows, there is \emph{not} a ``one to one''
85 relationship between revisions in the changelog, manifest, or filelog.
86 If the manifest hasn't changed between two changesets, the changelog
87 entries for those changesets will point to the same revision of the
88 manifest. If a file that Mercurial tracks hasn't changed between two
89 changesets, the entry for that file in the two revisions of the
90 manifest will point to the same revision of its filelog.
92 \section{Safe, efficient storage}
94 The underpinnings of changelogs, manifests, and filelogs are provided
95 by a single structure called the \emph{revlog}.
97 \subsection{Efficient storage}
99 The revlog provides efficient storage of revisions using a
100 \emph{delta} mechanism. Instead of storing a complete copy of a file
101 for each revision, it stores the changes needed to transform an older
102 revision into the new revision. For many kinds of file data, these
103 deltas are typically a fraction of a percent of the size of a full
104 copy of a file.
106 Some obsolete revision control systems can only work with deltas of
107 text files. They must either store binary files as complete snapshots
108 or encoded into a text representation, both of which are wasteful
109 approaches. Mercurial can efficiently handle deltas of files with
110 arbitrary binary contents; it doesn't need to treat text as special.
112 \subsection{Safe operation}
113 \label{sec:concepts:txn}
115 Mercurial only ever \emph{appends} data to the end of a revlog file.
116 It never modifies a section of a file after it has written it. This
117 is both more robust and efficient than schemes that need to modify or
118 rewrite data.
120 In addition, Mercurial treats every write as part of a
121 \emph{transaction} that can span a number of files. A transaction is
122 \emph{atomic}: either the entire transaction succeeds and its effects
123 are all visible to readers in one go, or the whole thing is undone.
124 This guarantee of atomicity means that if you're running two copies of
125 Mercurial, where one is reading data and one is writing it, the reader
126 will never see a partially written result that might confuse it.
128 The fact that Mercurial only appends to files makes it easier to
129 provide this transactional guarantee. The easier it is to do stuff
130 like this, the more confident you should be that it's done correctly.
132 \subsection{Fast retrieval}
134 Mercurial cleverly avoids a pitfall common to all earlier
135 revision control systems: the problem of \emph{inefficient retrieval}.
136 Most revision control systems store the contents of a revision as an
137 incremental series of modifications against a ``snapshot''. To
138 reconstruct a specific revision, you must first read the snapshot, and
139 then every one of the revisions between the snapshot and your target
140 revision. The more history that a file accumulates, the more
141 revisions you must read, hence the longer it takes to reconstruct a
142 particular revision.
144 \begin{figure}[ht]
145 \centering
146 \grafix{snapshot}
147 \caption{Snapshot of a revlog, with incremental deltas}
148 \label{fig:concepts:snapshot}
149 \end{figure}
151 The innovation that Mercurial applies to this problem is simple but
152 effective. Once the cumulative amount of delta information stored
153 since the last snapshot exceeds a fixed threshold, it stores a new
154 snapshot (compressed, of course), instead of another delta. This
155 makes it possible to reconstruct \emph{any} revision of a file
156 quickly. This approach works so well that it has since been copied by
157 several other revision control systems.
159 Figure~\ref{fig:concepts:snapshot} illustrates the idea. In an entry
160 in a revlog's index file, Mercurial stores the range of entries from
161 the data file that it must read to reconstruct a particular revision.
163 \subsubsection{Aside: the influence of video compression}
165 If you're familiar with video compression or have ever watched a TV
166 feed through a digital cable or satellite service, you may know that
167 most video compression schemes store each frame of video as a delta
168 against its predecessor frame. In addition, these schemes use
169 ``lossy'' compression techniques to increase the compression ratio, so
170 visual errors accumulate over the course of a number of inter-frame
171 deltas.
173 Because it's possible for a video stream to ``drop out'' occasionally
174 due to signal glitches, and to limit the accumulation of artefacts
175 introduced by the lossy compression process, video encoders
176 periodically insert a complete frame (called a ``key frame'') into the
177 video stream; the next delta is generated against that frame. This
178 means that if the video signal gets interrupted, it will resume once
179 the next key frame is received. Also, the accumulation of encoding
180 errors restarts anew with each key frame.
182 \subsection{Identification and strong integrity}
184 Along with delta or snapshot information, a revlog entry contains a
185 cryptographic hash of the data that it represents. This makes it
186 difficult to forge the contents of a revision, and easy to detect
187 accidental corruption.
189 Hashes provide more than a mere check against corruption; they are
190 used as the identifiers for revisions. The changeset identification
191 hashes that you see as an end user are from revisions of the
192 changelog. Although filelogs and the manifest also use hashes,
193 Mercurial only uses these behind the scenes.
195 Mercurial verifies that hashes are correct when it retrieves file
196 revisions and when it pulls changes from another repository. If it
197 encounters an integrity problem, it will complain and stop whatever
198 it's doing.
200 In addition to the effect it has on retrieval efficiency, Mercurial's
201 use of periodic snapshots makes it more robust against partial data
202 corruption. If a revlog becomes partly corrupted due to a hardware
203 error or system bug, it's often possible to reconstruct some or most
204 revisions from the uncorrupted sections of the revlog, both before and
205 after the corrupted section. This would not be possible with a
206 delta-only storage model.
208 \section{Revision history, branching,
209 and merging}
211 Every entry in a Mercurial revlog knows the identity of its immediate
212 ancestor revision, usually referred to as its \emph{parent}. In fact,
213 a revision contains room for not one parent, but two. Mercurial uses
214 a special hash, called the ``null ID'', to represent the idea ``there
215 is no parent here''. This hash is simply a string of zeroes.
217 In figure~\ref{fig:concepts:revlog}, you can see an example of the
218 conceptual structure of a revlog. Filelogs, manifests, and changelogs
219 all have this same structure; they differ only in the kind of data
220 stored in each delta or snapshot.
222 The first revision in a revlog (at the bottom of the image) has the
223 null ID in both of its parent slots. For a ``normal'' revision, its
224 first parent slot contains the ID of its parent revision, and its
225 second contains the null ID, indicating that the revision has only one
226 real parent. Any two revisions that have the same parent ID are
227 branches. A revision that represents a merge between branches has two
228 normal revision IDs in its parent slots.
230 \begin{figure}[ht]
231 \centering
232 \grafix{revlog}
233 \caption{}
234 \label{fig:concepts:revlog}
235 \end{figure}
237 \section{The working directory}
239 In the working directory, Mercurial stores a snapshot of the files
240 from the repository as of a particular changeset.
242 The working directory ``knows'' which changeset it contains. When you
243 update the working directory to contain a particular changeset,
244 Mercurial looks up the appropriate revision of the manifest to find
245 out which files it was tracking at the time that changeset was
246 committed, and which revision of each file was then current. It then
247 recreates a copy of each of those files, with the same contents it had
248 when the changeset was committed.
250 The \emph{dirstate} contains Mercurial's knowledge of the working
251 directory. This details which changeset the working directory is
252 updated to, and all of the files that Mercurial is tracking in the
253 working directory.
255 Just as a revision of a revlog has room for two parents, so that it
256 can represent either a normal revision (with one parent) or a merge of
257 two earlier revisions, the dirstate has slots for two parents. When
258 you use the \hgcmd{update} command, the changeset that you update to
259 is stored in the ``first parent'' slot, and the null ID in the second.
260 When you \hgcmd{merge} with another changeset, the first parent
261 remains unchanged, and the second parent is filled in with the
262 changeset you're merging with. The \hgcmd{parents} command tells you
263 what the parents of the dirstate are.
265 \subsection{What happens when you commit}
267 The dirstate stores parent information for more than just book-keeping
268 purposes. Mercurial uses the parents of the dirstate as \emph{the
269 parents of a new changeset} when you perform a commit.
271 \begin{figure}[ht]
272 \centering
273 \grafix{wdir}
274 \caption{The working directory can have two parents}
275 \label{fig:concepts:wdir}
276 \end{figure}
278 Figure~\ref{fig:concepts:wdir} shows the normal state of the working
279 directory, where it has a single changeset as parent. That changeset
280 is the \emph{tip}, the newest changeset in the repository that has no
281 children.
283 \begin{figure}[ht]
284 \centering
285 \grafix{wdir-after-commit}
286 \caption{The working directory gains new parents after a commit}
287 \label{fig:concepts:wdir-after-commit}
288 \end{figure}
290 It's useful to think of the working directory as ``the changeset I'm
291 about to commit''. Any files that you tell Mercurial that you've
292 added, removed, renamed, or copied will be reflected in that
293 changeset, as will modifications to any files that Mercurial is
294 already tracking; the new changeset will have the parents of the
295 working directory as its parents.
297 After a commit, Mercurial will update the parents of the working
298 directory, so that the first parent is the ID of the new changeset,
299 and the second is the null ID. This is shown in
300 figure~\ref{fig:concepts:wdir-after-commit}. Mercurial doesn't touch
301 any of the files in the working directory when you commit; it just
302 modifies the dirstate to note its new parents.
304 \subsection{Creating a new head}
306 It's perfectly normal to update the working directory to a changeset
307 other than the current tip. For example, you might want to know what
308 your project looked like last Tuesday, or you could be looking through
309 changesets to see which one introduced a bug. In cases like this, the
310 natural thing to do is update the working directory to the changeset
311 you're interested in, and then examine the files in the working
312 directory directly to see their contents as they werea when you
313 committed that changeset. The effect of this is shown in
314 figure~\ref{fig:concepts:wdir-pre-branch}.
316 \begin{figure}[ht]
317 \centering
318 \grafix{wdir-pre-branch}
319 \caption{The working directory, updated to an older changeset}
320 \label{fig:concepts:wdir-pre-branch}
321 \end{figure}
323 Having updated the working directory to an older changeset, what
324 happens if you make some changes, and then commit? Mercurial behaves
325 in the same way as I outlined above. The parents of the working
326 directory become the parents of the new changeset. This new changeset
327 has no children, so it becomes the new tip. And the repository now
328 contains two changesets that have no children; we call these
329 \emph{heads}. You can see the structure that this creates in
330 figure~\ref{fig:concepts:wdir-branch}.
332 \begin{figure}[ht]
333 \centering
334 \grafix{wdir-branch}
335 \caption{After a commit made while synced to an older changeset}
336 \label{fig:concepts:wdir-branch}
337 \end{figure}
339 \begin{note}
340 If you're new to Mercurial, you should keep in mind a common
341 ``error'', which is to use the \hgcmd{pull} command without any
342 options. By default, the \hgcmd{pull} command \emph{does not}
343 update the working directory, so you'll bring new changesets into
344 your repository, but the working directory will stay synced at the
345 same changeset as before the pull. If you make some changes and
346 commit afterwards, you'll thus create a new head, because your
347 working directory isn't synced to whatever the current tip is.
349 I put the word ``error'' in quotes because all that you need to do
350 to rectify this situation is \hgcmd{merge}, then \hgcmd{commit}. In
351 other words, this almost never has negative consequences; it just
352 surprises people. I'll discuss other ways to avoid this behaviour,
353 and why Mercurial behaves in this initially surprising way, later
354 on.
355 \end{note}
357 \subsection{Merging heads}
359 When you run the \hgcmd{merge} command, Mercurial leaves the first
360 parent of the working directory unchanged, and sets the second parent
361 to the changeset you're merging with, as shown in
362 figure~\ref{fig:concepts:wdir-merge}.
364 \begin{figure}[ht]
365 \centering
366 \grafix{wdir-merge}
367 \caption{Merging two heads}
368 \label{fig:concepts:wdir-merge}
369 \end{figure}
371 Mercurial also has to modify the working directory, to merge the files
372 managed in the two changesets. Simplified a little, the merging
373 process goes like this, for every file in the manifests of both
374 changesets.
375 \begin{itemize}
376 \item If neither changeset has modified a file, do nothing with that
377 file.
378 \item If one changeset has modified a file, and the other hasn't,
379 create the modified copy of the file in the working directory.
380 \item If one changeset has removed a file, and the other hasn't (or
381 has also deleted it), delete the file from the working directory.
382 \item If one changeset has removed a file, but the other has modified
383 the file, ask the user what to do: keep the modified file, or remove
384 it?
385 \item If both changesets have modified a file, invoke an external
386 merge program to choose the new contents for the merged file. This
387 may require input from the user.
388 \item If one changeset has modified a file, and the other has renamed
389 or copied the file, make sure that the changes follow the new name
390 of the file.
391 \end{itemize}
392 There are more details---merging has plenty of corner cases---but
393 these are the most common choices that are involved in a merge. As
394 you can see, most cases are completely automatic, and indeed most
395 merges finish automatically, without requiring your input to resolve
396 any conflicts.
398 When you're thinking about what happens when you commit after a merge,
399 once again the working directory is ``the changeset I'm about to
400 commit''. After the \hgcmd{merge} command completes, the working
401 directory has two parents; these will become the parents of the new
402 changeset.
404 Mercurial lets you perform multiple merges, but you must commit the
405 results of each individual merge as you go. This is necessary because
406 Mercurial only tracks two parents for both revisions and the working
407 directory. While it would be technically possible to merge multiple
408 changesets at once, the prospect of user confusion and making a
409 terrible mess of a merge immediately becomes overwhelming.
411 \section{Other interesting design features}
413 In the sections above, I've tried to highlight some of the most
414 important aspects of Mercurial's design, to illustrate that it pays
415 careful attention to reliability and performance. However, the
416 attention to detail doesn't stop there. There are a number of other
417 aspects of Mercurial's construction that I personally find
418 interesting. I'll detail a few of them here, separate from the ``big
419 ticket'' items above, so that if you're interested, you can gain a
420 better idea of the amount of thinking that goes into a well-designed
421 system.
423 \subsection{Clever compression}
425 When appropriate, Mercurial will store both snapshots and deltas in
426 compressed form. It does this by always \emph{trying to} compress a
427 snapshot or delta, but only storing the compressed version if it's
428 smaller than the uncompressed version.
430 This means that Mercurial does ``the right thing'' when storing a file
431 whose native form is compressed, such as a \texttt{zip} archive or a
432 JPEG image. When these types of files are compressed a second time,
433 the resulting file is usually bigger than the once-compressed form,
434 and so Mercurial will store the plain \texttt{zip} or JPEG.
436 Deltas between revisions of a compressed file are usually larger than
437 snapshots of the file, and Mercurial again does ``the right thing'' in
438 these cases. It finds that such a delta exceeds the threshold at
439 which it should store a complete snapshot of the file, so it stores
440 the snapshot, again saving space compared to a naive delta-only
441 approach.
443 \subsubsection{Network recompression}
445 When storing revisions on disk, Mercurial uses the ``deflate''
446 compression algorithm (the same one used by the popular \texttt{zip}
447 archive format), which balances good speed with a respectable
448 compression ratio. However, when transmitting revision data over a
449 network connection, Mercurial uncompresses the compressed revision
450 data.
452 If the connection is over HTTP, Mercurial recompresses the entire
453 stream of data using a compression algorithm that gives a better
454 compression ratio (the Burrows-Wheeler algorithm from the widely used
455 \texttt{bzip2} compression package). This combination of algorithm
456 and compression of the entire stream (instead of a revision at a time)
457 substantially reduces the number of bytes to be transferred, yielding
458 better network performance over almost all kinds of network.
460 (If the connection is over \command{ssh}, Mercurial \emph{doesn't}
461 recompress the stream, because \command{ssh} can already do this
462 itself.)
464 \subsection{Read/write ordering and atomicity}
466 Appending to files isn't the whole story when it comes to guaranteeing
467 that a reader won't see a partial write. If you recall
468 figure~\ref{fig:concepts:metadata}, revisions in the changelog point to
469 revisions in the manifest, and revisions in the manifest point to
470 revisions in filelogs. This hierarchy is deliberate.
472 A writer starts a transaction by writing filelog and manifest data,
473 and doesn't write any changelog data until those are finished. A
474 reader starts by reading changelog data, then manifest data, followed
475 by filelog data.
477 Since the writer has always finished writing filelog and manifest data
478 before it writes to the changelog, a reader will never read a pointer
479 to a partially written manifest revision from the changelog, and it will
480 never read a pointer to a partially written filelog revision from the
481 manifest.
483 \subsection{Concurrent access}
485 The read/write ordering and atomicity guarantees mean that Mercurial
486 never needs to \emph{lock} a repository when it's reading data, even
487 if the repository is being written to while the read is occurring.
488 This has a big effect on scalability; you can have an arbitrary number
489 of Mercurial processes safely reading data from a repository safely
490 all at once, no matter whether it's being written to or not.
492 The lockless nature of reading means that if you're sharing a
493 repository on a multi-user system, you don't need to grant other local
494 users permission to \emph{write} to your repository in order for them
495 to be able to clone it or pull changes from it; they only need
496 \emph{read} permission. (This is \emph{not} a common feature among
497 revision control systems, so don't take it for granted! Most require
498 readers to be able to lock a repository to access it safely, and this
499 requires write permission on at least one directory, which of course
500 makes for all kinds of nasty and annoying security and administrative
501 problems.)
503 Mercurial uses locks to ensure that only one process can write to a
504 repository at a time (the locking mechanism is safe even over
505 filesystems that are notoriously hostile to locking, such as NFS). If
506 a repository is locked, a writer will wait for a while to retry if the
507 repository becomes unlocked, but if the repository remains locked for
508 too long, the process attempting to write will time out after a while.
509 This means that your daily automated scripts won't get stuck forever
510 and pile up if a system crashes unnoticed, for example. (Yes, the
511 timeout is configurable, from zero to infinity.)
513 \subsubsection{Safe dirstate access}
515 As with revision data, Mercurial doesn't take a lock to read the
516 dirstate file; it does acquire a lock to write it. To avoid the
517 possibility of reading a partially written copy of the dirstate file,
518 Mercurial writes to a file with a unique name in the same directory as
519 the dirstate file, then renames the temporary file atomically to
520 \filename{dirstate}. The file named \filename{dirstate} is thus
521 guaranteed to be complete, not partially written.
523 \subsection{Avoiding seeks}
525 Critical to Mercurial's performance is the avoidance of seeks of the
526 disk head, since any seek is far more expensive than even a
527 comparatively large read operation.
529 This is why, for example, the dirstate is stored in a single file. If
530 there were a dirstate file per directory that Mercurial tracked, the
531 disk would seek once per directory. Instead, Mercurial reads the
532 entire single dirstate file in one step.
534 Mercurial also uses a ``copy on write'' scheme when cloning a
535 repository on local storage. Instead of copying every revlog file
536 from the old repository into the new repository, it makes a ``hard
537 link'', which is a shorthand way to say ``these two names point to the
538 same file''. When Mercurial is about to write to one of a revlog's
539 files, it checks to see if the number of names pointing at the file is
540 greater than one. If it is, more than one repository is using the
541 file, so Mercurial makes a new copy of the file that is private to
542 this repository.
544 A few revision control developers have pointed out that this idea of
545 making a complete private copy of a file is not very efficient in its
546 use of storage. While this is true, storage is cheap, and this method
547 gives the highest performance while deferring most book-keeping to the
548 operating system. An alternative scheme would most likely reduce
549 performance and increase the complexity of the software, each of which
550 is much more important to the ``feel'' of day-to-day use.
552 \subsection{Other contents of the dirstate}
554 Because Mercurial doesn't force you to tell it when you're modifying a
555 file, it uses the dirstate to store some extra information so it can
556 determine efficiently whether you have modified a file. For each file
557 in the working directory, it stores the time that it last modified the
558 file itself, and the size of the file at that time.
560 When you explicitly \hgcmd{add}, \hgcmd{remove}, \hgcmd{rename} or
561 \hgcmd{copy} files, Mercurial updates the dirstate so that it knows
562 what to do with those files when you commit.
564 When Mercurial is checking the states of files in the working
565 directory, it first checks a file's modification time. If that has
566 not changed, the file must not have been modified. If the file's size
567 has changed, the file must have been modified. If the modification
568 time has changed, but the size has not, only then does Mercurial need
569 to read the actual contents of the file to see if they've changed.
570 Storing these few extra pieces of information dramatically reduces the
571 amount of data that Mercurial needs to read, which yields large
572 performance improvements compared to other revision control systems.
574 %%% Local Variables:
575 %%% mode: latex
576 %%% TeX-master: "00book"
577 %%% End: