hgbook
view es/concepts.tex @ 395:149ea8ae39c4
translated a couple of paragraphs of the "behind the scenes" chapter
| author | Javier Rojas <jerojasro@devnull.li> |
|---|---|
| date | Tue Nov 04 23:50:20 2008 -0500 (2008-11-04) |
| parents | 2c2c86824c61 |
| children | 0f2170fcb79a |
line source
1 \chapter{Tras bambalinas}
2 \label{chap:concepts}
4 A diferencia de varios sistemas de control de revisiones, los
5 conceptos en los que se fundamenta Mercurial son lo suficientemente
6 simples como para entender fácilmente cómo funciona el software.
7 Saber esto no es necesario, pero considero útil tener un ``modelo
8 mental'' de qué es lo que sucede.
10 Comprender esto me da la confianza de que Mercurial ha sido
11 cuidadosamente diseñado para ser tanto \emph{seguro} como
12 \emph{eficiente}. Y tal vez con la misma importancia, si es fácil
13 para mí hacerme a una idea adecuada de qué está haciendo el software
14 cuando llevo a cabo una tarea relacionada con control de revisiones,
15 es menos probable que me sosprenda su comportamiento.
17 En este capítulo, cubriremos inicialmente los conceptos centrales
18 del diseño de Mercurial, y luego discutiremos algunos detalles
19 interesantes de su implementación.
21 \section{Registro del historial de Mercurial}
23 \subsection{Seguir el historial de un único fichero}
25 Cuando Mercurial sigue las modificaciones a un fichero, guarda el
26 historial de dicho fichero en un objeto de metadatos llamado
27 \emph{filelog}\ndt{Fichero de registro}. Cada entrada en el fichero
28 de registro contiene suficiente información para reconstruir una
29 revisión del fichero que se está siguiendo. Los ficheros de registro
30 son almacenados como ficheros el el directorio
31 \sdirname{.hg/store/data}. Un fichero de registro contiene dos tipos
32 de información: datos de revisiones, y un índice para ayudar a
33 Mercurial a buscar revisiones eficientemente.
35 El fichero de registro de un fichero grande, o con un historial muy
36 largo, es guardado como ficheros separados para datos (sufijo
37 ``\texttt{.d}'') y para el índice (sufijo ``\texttt{.i}''). Para
38 ficheros pequeños con un historial pequeño, los datos de revisiones y
39 el índice son combinados en un único fichero ``\texttt{.i}''. La
40 correspondencia entre un fichero en el directorio de trabajo y el
41 fichero de registro que hace seguimiento a su historial en el
42 repositorio se ilustra en la figura~\ref{fig:concepts:filelog}.
44 \begin{figure}[ht]
45 \centering
46 \grafix{filelog}
47 \caption{Relationships between files in working directory and
48 filelogs in repository}
49 \label{fig:concepts:filelog}
50 \end{figure}
52 \subsection{Managing tracked files}
54 Mercurial uses a structure called a \emph{manifest} to collect
55 together information about the files that it tracks. Each entry in
56 the manifest contains information about the files present in a single
57 changeset. An entry records which files are present in the changeset,
58 the revision of each file, and a few other pieces of file metadata.
60 \subsection{Recording changeset information}
62 The \emph{changelog} contains information about each changeset. Each
63 revision records who committed a change, the changeset comment, other
64 pieces of changeset-related information, and the revision of the
65 manifest to use.
67 \subsection{Relationships between revisions}
69 Within a changelog, a manifest, or a filelog, each revision stores a
70 pointer to its immediate parent (or to its two parents, if it's a
71 merge revision). As I mentioned above, there are also relationships
72 between revisions \emph{across} these structures, and they are
73 hierarchical in nature.
75 For every changeset in a repository, there is exactly one revision
76 stored in the changelog. Each revision of the changelog contains a
77 pointer to a single revision of the manifest. A revision of the
78 manifest stores a pointer to a single revision of each filelog tracked
79 when that changeset was created. These relationships are illustrated
80 in figure~\ref{fig:concepts:metadata}.
82 \begin{figure}[ht]
83 \centering
84 \grafix{metadata}
85 \caption{Metadata relationships}
86 \label{fig:concepts:metadata}
87 \end{figure}
89 As the illustration shows, there is \emph{not} a ``one to one''
90 relationship between revisions in the changelog, manifest, or filelog.
91 If the manifest hasn't changed between two changesets, the changelog
92 entries for those changesets will point to the same revision of the
93 manifest. If a file that Mercurial tracks hasn't changed between two
94 changesets, the entry for that file in the two revisions of the
95 manifest will point to the same revision of its filelog.
97 \section{Safe, efficient storage}
99 The underpinnings of changelogs, manifests, and filelogs are provided
100 by a single structure called the \emph{revlog}.
102 \subsection{Efficient storage}
104 The revlog provides efficient storage of revisions using a
105 \emph{delta} mechanism. Instead of storing a complete copy of a file
106 for each revision, it stores the changes needed to transform an older
107 revision into the new revision. For many kinds of file data, these
108 deltas are typically a fraction of a percent of the size of a full
109 copy of a file.
111 Some obsolete revision control systems can only work with deltas of
112 text files. They must either store binary files as complete snapshots
113 or encoded into a text representation, both of which are wasteful
114 approaches. Mercurial can efficiently handle deltas of files with
115 arbitrary binary contents; it doesn't need to treat text as special.
117 \subsection{Safe operation}
118 \label{sec:concepts:txn}
120 Mercurial only ever \emph{appends} data to the end of a revlog file.
121 It never modifies a section of a file after it has written it. This
122 is both more robust and efficient than schemes that need to modify or
123 rewrite data.
125 In addition, Mercurial treats every write as part of a
126 \emph{transaction} that can span a number of files. A transaction is
127 \emph{atomic}: either the entire transaction succeeds and its effects
128 are all visible to readers in one go, or the whole thing is undone.
129 This guarantee of atomicity means that if you're running two copies of
130 Mercurial, where one is reading data and one is writing it, the reader
131 will never see a partially written result that might confuse it.
133 The fact that Mercurial only appends to files makes it easier to
134 provide this transactional guarantee. The easier it is to do stuff
135 like this, the more confident you should be that it's done correctly.
137 \subsection{Fast retrieval}
139 Mercurial cleverly avoids a pitfall common to all earlier
140 revision control systems: the problem of \emph{inefficient retrieval}.
141 Most revision control systems store the contents of a revision as an
142 incremental series of modifications against a ``snapshot''. To
143 reconstruct a specific revision, you must first read the snapshot, and
144 then every one of the revisions between the snapshot and your target
145 revision. The more history that a file accumulates, the more
146 revisions you must read, hence the longer it takes to reconstruct a
147 particular revision.
149 \begin{figure}[ht]
150 \centering
151 \grafix{snapshot}
152 \caption{Snapshot of a revlog, with incremental deltas}
153 \label{fig:concepts:snapshot}
154 \end{figure}
156 The innovation that Mercurial applies to this problem is simple but
157 effective. Once the cumulative amount of delta information stored
158 since the last snapshot exceeds a fixed threshold, it stores a new
159 snapshot (compressed, of course), instead of another delta. This
160 makes it possible to reconstruct \emph{any} revision of a file
161 quickly. This approach works so well that it has since been copied by
162 several other revision control systems.
164 Figure~\ref{fig:concepts:snapshot} illustrates the idea. In an entry
165 in a revlog's index file, Mercurial stores the range of entries from
166 the data file that it must read to reconstruct a particular revision.
168 \subsubsection{Aside: the influence of video compression}
170 If you're familiar with video compression or have ever watched a TV
171 feed through a digital cable or satellite service, you may know that
172 most video compression schemes store each frame of video as a delta
173 against its predecessor frame. In addition, these schemes use
174 ``lossy'' compression techniques to increase the compression ratio, so
175 visual errors accumulate over the course of a number of inter-frame
176 deltas.
178 Because it's possible for a video stream to ``drop out'' occasionally
179 due to signal glitches, and to limit the accumulation of artefacts
180 introduced by the lossy compression process, video encoders
181 periodically insert a complete frame (called a ``key frame'') into the
182 video stream; the next delta is generated against that frame. This
183 means that if the video signal gets interrupted, it will resume once
184 the next key frame is received. Also, the accumulation of encoding
185 errors restarts anew with each key frame.
187 \subsection{Identification and strong integrity}
189 Along with delta or snapshot information, a revlog entry contains a
190 cryptographic hash of the data that it represents. This makes it
191 difficult to forge the contents of a revision, and easy to detect
192 accidental corruption.
194 Hashes provide more than a mere check against corruption; they are
195 used as the identifiers for revisions. The changeset identification
196 hashes that you see as an end user are from revisions of the
197 changelog. Although filelogs and the manifest also use hashes,
198 Mercurial only uses these behind the scenes.
200 Mercurial verifies that hashes are correct when it retrieves file
201 revisions and when it pulls changes from another repository. If it
202 encounters an integrity problem, it will complain and stop whatever
203 it's doing.
205 In addition to the effect it has on retrieval efficiency, Mercurial's
206 use of periodic snapshots makes it more robust against partial data
207 corruption. If a revlog becomes partly corrupted due to a hardware
208 error or system bug, it's often possible to reconstruct some or most
209 revisions from the uncorrupted sections of the revlog, both before and
210 after the corrupted section. This would not be possible with a
211 delta-only storage model.
213 \section{Revision history, branching,
214 and merging}
216 Every entry in a Mercurial revlog knows the identity of its immediate
217 ancestor revision, usually referred to as its \emph{parent}. In fact,
218 a revision contains room for not one parent, but two. Mercurial uses
219 a special hash, called the ``null ID'', to represent the idea ``there
220 is no parent here''. This hash is simply a string of zeroes.
222 In figure~\ref{fig:concepts:revlog}, you can see an example of the
223 conceptual structure of a revlog. Filelogs, manifests, and changelogs
224 all have this same structure; they differ only in the kind of data
225 stored in each delta or snapshot.
227 The first revision in a revlog (at the bottom of the image) has the
228 null ID in both of its parent slots. For a ``normal'' revision, its
229 first parent slot contains the ID of its parent revision, and its
230 second contains the null ID, indicating that the revision has only one
231 real parent. Any two revisions that have the same parent ID are
232 branches. A revision that represents a merge between branches has two
233 normal revision IDs in its parent slots.
235 \begin{figure}[ht]
236 \centering
237 \grafix{revlog}
238 \caption{}
239 \label{fig:concepts:revlog}
240 \end{figure}
242 \section{The working directory}
244 In the working directory, Mercurial stores a snapshot of the files
245 from the repository as of a particular changeset.
247 The working directory ``knows'' which changeset it contains. When you
248 update the working directory to contain a particular changeset,
249 Mercurial looks up the appropriate revision of the manifest to find
250 out which files it was tracking at the time that changeset was
251 committed, and which revision of each file was then current. It then
252 recreates a copy of each of those files, with the same contents it had
253 when the changeset was committed.
255 The \emph{dirstate} contains Mercurial's knowledge of the working
256 directory. This details which changeset the working directory is
257 updated to, and all of the files that Mercurial is tracking in the
258 working directory.
260 Just as a revision of a revlog has room for two parents, so that it
261 can represent either a normal revision (with one parent) or a merge of
262 two earlier revisions, the dirstate has slots for two parents. When
263 you use the \hgcmd{update} command, the changeset that you update to
264 is stored in the ``first parent'' slot, and the null ID in the second.
265 When you \hgcmd{merge} with another changeset, the first parent
266 remains unchanged, and the second parent is filled in with the
267 changeset you're merging with. The \hgcmd{parents} command tells you
268 what the parents of the dirstate are.
270 \subsection{What happens when you commit}
272 The dirstate stores parent information for more than just book-keeping
273 purposes. Mercurial uses the parents of the dirstate as \emph{the
274 parents of a new changeset} when you perform a commit.
276 \begin{figure}[ht]
277 \centering
278 \grafix{wdir}
279 \caption{The working directory can have two parents}
280 \label{fig:concepts:wdir}
281 \end{figure}
283 Figure~\ref{fig:concepts:wdir} shows the normal state of the working
284 directory, where it has a single changeset as parent. That changeset
285 is the \emph{tip}, the newest changeset in the repository that has no
286 children.
288 \begin{figure}[ht]
289 \centering
290 \grafix{wdir-after-commit}
291 \caption{The working directory gains new parents after a commit}
292 \label{fig:concepts:wdir-after-commit}
293 \end{figure}
295 It's useful to think of the working directory as ``the changeset I'm
296 about to commit''. Any files that you tell Mercurial that you've
297 added, removed, renamed, or copied will be reflected in that
298 changeset, as will modifications to any files that Mercurial is
299 already tracking; the new changeset will have the parents of the
300 working directory as its parents.
302 After a commit, Mercurial will update the parents of the working
303 directory, so that the first parent is the ID of the new changeset,
304 and the second is the null ID. This is shown in
305 figure~\ref{fig:concepts:wdir-after-commit}. Mercurial doesn't touch
306 any of the files in the working directory when you commit; it just
307 modifies the dirstate to note its new parents.
309 \subsection{Creating a new head}
311 It's perfectly normal to update the working directory to a changeset
312 other than the current tip. For example, you might want to know what
313 your project looked like last Tuesday, or you could be looking through
314 changesets to see which one introduced a bug. In cases like this, the
315 natural thing to do is update the working directory to the changeset
316 you're interested in, and then examine the files in the working
317 directory directly to see their contents as they werea when you
318 committed that changeset. The effect of this is shown in
319 figure~\ref{fig:concepts:wdir-pre-branch}.
321 \begin{figure}[ht]
322 \centering
323 \grafix{wdir-pre-branch}
324 \caption{The working directory, updated to an older changeset}
325 \label{fig:concepts:wdir-pre-branch}
326 \end{figure}
328 Having updated the working directory to an older changeset, what
329 happens if you make some changes, and then commit? Mercurial behaves
330 in the same way as I outlined above. The parents of the working
331 directory become the parents of the new changeset. This new changeset
332 has no children, so it becomes the new tip. And the repository now
333 contains two changesets that have no children; we call these
334 \emph{heads}. You can see the structure that this creates in
335 figure~\ref{fig:concepts:wdir-branch}.
337 \begin{figure}[ht]
338 \centering
339 \grafix{wdir-branch}
340 \caption{After a commit made while synced to an older changeset}
341 \label{fig:concepts:wdir-branch}
342 \end{figure}
344 \begin{note}
345 If you're new to Mercurial, you should keep in mind a common
346 ``error'', which is to use the \hgcmd{pull} command without any
347 options. By default, the \hgcmd{pull} command \emph{does not}
348 update the working directory, so you'll bring new changesets into
349 your repository, but the working directory will stay synced at the
350 same changeset as before the pull. If you make some changes and
351 commit afterwards, you'll thus create a new head, because your
352 working directory isn't synced to whatever the current tip is.
354 I put the word ``error'' in quotes because all that you need to do
355 to rectify this situation is \hgcmd{merge}, then \hgcmd{commit}. In
356 other words, this almost never has negative consequences; it just
357 surprises people. I'll discuss other ways to avoid this behaviour,
358 and why Mercurial behaves in this initially surprising way, later
359 on.
360 \end{note}
362 \subsection{Merging heads}
364 When you run the \hgcmd{merge} command, Mercurial leaves the first
365 parent of the working directory unchanged, and sets the second parent
366 to the changeset you're merging with, as shown in
367 figure~\ref{fig:concepts:wdir-merge}.
369 \begin{figure}[ht]
370 \centering
371 \grafix{wdir-merge}
372 \caption{Merging two heads}
373 \label{fig:concepts:wdir-merge}
374 \end{figure}
376 Mercurial also has to modify the working directory, to merge the files
377 managed in the two changesets. Simplified a little, the merging
378 process goes like this, for every file in the manifests of both
379 changesets.
380 \begin{itemize}
381 \item If neither changeset has modified a file, do nothing with that
382 file.
383 \item If one changeset has modified a file, and the other hasn't,
384 create the modified copy of the file in the working directory.
385 \item If one changeset has removed a file, and the other hasn't (or
386 has also deleted it), delete the file from the working directory.
387 \item If one changeset has removed a file, but the other has modified
388 the file, ask the user what to do: keep the modified file, or remove
389 it?
390 \item If both changesets have modified a file, invoke an external
391 merge program to choose the new contents for the merged file. This
392 may require input from the user.
393 \item If one changeset has modified a file, and the other has renamed
394 or copied the file, make sure that the changes follow the new name
395 of the file.
396 \end{itemize}
397 There are more details---merging has plenty of corner cases---but
398 these are the most common choices that are involved in a merge. As
399 you can see, most cases are completely automatic, and indeed most
400 merges finish automatically, without requiring your input to resolve
401 any conflicts.
403 When you're thinking about what happens when you commit after a merge,
404 once again the working directory is ``the changeset I'm about to
405 commit''. After the \hgcmd{merge} command completes, the working
406 directory has two parents; these will become the parents of the new
407 changeset.
409 Mercurial lets you perform multiple merges, but you must commit the
410 results of each individual merge as you go. This is necessary because
411 Mercurial only tracks two parents for both revisions and the working
412 directory. While it would be technically possible to merge multiple
413 changesets at once, the prospect of user confusion and making a
414 terrible mess of a merge immediately becomes overwhelming.
416 \section{Other interesting design features}
418 In the sections above, I've tried to highlight some of the most
419 important aspects of Mercurial's design, to illustrate that it pays
420 careful attention to reliability and performance. However, the
421 attention to detail doesn't stop there. There are a number of other
422 aspects of Mercurial's construction that I personally find
423 interesting. I'll detail a few of them here, separate from the ``big
424 ticket'' items above, so that if you're interested, you can gain a
425 better idea of the amount of thinking that goes into a well-designed
426 system.
428 \subsection{Clever compression}
430 When appropriate, Mercurial will store both snapshots and deltas in
431 compressed form. It does this by always \emph{trying to} compress a
432 snapshot or delta, but only storing the compressed version if it's
433 smaller than the uncompressed version.
435 This means that Mercurial does ``the right thing'' when storing a file
436 whose native form is compressed, such as a \texttt{zip} archive or a
437 JPEG image. When these types of files are compressed a second time,
438 the resulting file is usually bigger than the once-compressed form,
439 and so Mercurial will store the plain \texttt{zip} or JPEG.
441 Deltas between revisions of a compressed file are usually larger than
442 snapshots of the file, and Mercurial again does ``the right thing'' in
443 these cases. It finds that such a delta exceeds the threshold at
444 which it should store a complete snapshot of the file, so it stores
445 the snapshot, again saving space compared to a naive delta-only
446 approach.
448 \subsubsection{Network recompression}
450 When storing revisions on disk, Mercurial uses the ``deflate''
451 compression algorithm (the same one used by the popular \texttt{zip}
452 archive format), which balances good speed with a respectable
453 compression ratio. However, when transmitting revision data over a
454 network connection, Mercurial uncompresses the compressed revision
455 data.
457 If the connection is over HTTP, Mercurial recompresses the entire
458 stream of data using a compression algorithm that gives a better
459 compression ratio (the Burrows-Wheeler algorithm from the widely used
460 \texttt{bzip2} compression package). This combination of algorithm
461 and compression of the entire stream (instead of a revision at a time)
462 substantially reduces the number of bytes to be transferred, yielding
463 better network performance over almost all kinds of network.
465 (If the connection is over \command{ssh}, Mercurial \emph{doesn't}
466 recompress the stream, because \command{ssh} can already do this
467 itself.)
469 \subsection{Read/write ordering and atomicity}
471 Appending to files isn't the whole story when it comes to guaranteeing
472 that a reader won't see a partial write. If you recall
473 figure~\ref{fig:concepts:metadata}, revisions in the changelog point to
474 revisions in the manifest, and revisions in the manifest point to
475 revisions in filelogs. This hierarchy is deliberate.
477 A writer starts a transaction by writing filelog and manifest data,
478 and doesn't write any changelog data until those are finished. A
479 reader starts by reading changelog data, then manifest data, followed
480 by filelog data.
482 Since the writer has always finished writing filelog and manifest data
483 before it writes to the changelog, a reader will never read a pointer
484 to a partially written manifest revision from the changelog, and it will
485 never read a pointer to a partially written filelog revision from the
486 manifest.
488 \subsection{Concurrent access}
490 The read/write ordering and atomicity guarantees mean that Mercurial
491 never needs to \emph{lock} a repository when it's reading data, even
492 if the repository is being written to while the read is occurring.
493 This has a big effect on scalability; you can have an arbitrary number
494 of Mercurial processes safely reading data from a repository safely
495 all at once, no matter whether it's being written to or not.
497 The lockless nature of reading means that if you're sharing a
498 repository on a multi-user system, you don't need to grant other local
499 users permission to \emph{write} to your repository in order for them
500 to be able to clone it or pull changes from it; they only need
501 \emph{read} permission. (This is \emph{not} a common feature among
502 revision control systems, so don't take it for granted! Most require
503 readers to be able to lock a repository to access it safely, and this
504 requires write permission on at least one directory, which of course
505 makes for all kinds of nasty and annoying security and administrative
506 problems.)
508 Mercurial uses locks to ensure that only one process can write to a
509 repository at a time (the locking mechanism is safe even over
510 filesystems that are notoriously hostile to locking, such as NFS). If
511 a repository is locked, a writer will wait for a while to retry if the
512 repository becomes unlocked, but if the repository remains locked for
513 too long, the process attempting to write will time out after a while.
514 This means that your daily automated scripts won't get stuck forever
515 and pile up if a system crashes unnoticed, for example. (Yes, the
516 timeout is configurable, from zero to infinity.)
518 \subsubsection{Safe dirstate access}
520 As with revision data, Mercurial doesn't take a lock to read the
521 dirstate file; it does acquire a lock to write it. To avoid the
522 possibility of reading a partially written copy of the dirstate file,
523 Mercurial writes to a file with a unique name in the same directory as
524 the dirstate file, then renames the temporary file atomically to
525 \filename{dirstate}. The file named \filename{dirstate} is thus
526 guaranteed to be complete, not partially written.
528 \subsection{Avoiding seeks}
530 Critical to Mercurial's performance is the avoidance of seeks of the
531 disk head, since any seek is far more expensive than even a
532 comparatively large read operation.
534 This is why, for example, the dirstate is stored in a single file. If
535 there were a dirstate file per directory that Mercurial tracked, the
536 disk would seek once per directory. Instead, Mercurial reads the
537 entire single dirstate file in one step.
539 Mercurial also uses a ``copy on write'' scheme when cloning a
540 repository on local storage. Instead of copying every revlog file
541 from the old repository into the new repository, it makes a ``hard
542 link'', which is a shorthand way to say ``these two names point to the
543 same file''. When Mercurial is about to write to one of a revlog's
544 files, it checks to see if the number of names pointing at the file is
545 greater than one. If it is, more than one repository is using the
546 file, so Mercurial makes a new copy of the file that is private to
547 this repository.
549 A few revision control developers have pointed out that this idea of
550 making a complete private copy of a file is not very efficient in its
551 use of storage. While this is true, storage is cheap, and this method
552 gives the highest performance while deferring most book-keeping to the
553 operating system. An alternative scheme would most likely reduce
554 performance and increase the complexity of the software, each of which
555 is much more important to the ``feel'' of day-to-day use.
557 \subsection{Other contents of the dirstate}
559 Because Mercurial doesn't force you to tell it when you're modifying a
560 file, it uses the dirstate to store some extra information so it can
561 determine efficiently whether you have modified a file. For each file
562 in the working directory, it stores the time that it last modified the
563 file itself, and the size of the file at that time.
565 When you explicitly \hgcmd{add}, \hgcmd{remove}, \hgcmd{rename} or
566 \hgcmd{copy} files, Mercurial updates the dirstate so that it knows
567 what to do with those files when you commit.
569 When Mercurial is checking the states of files in the working
570 directory, it first checks a file's modification time. If that has
571 not changed, the file must not have been modified. If the file's size
572 has changed, the file must have been modified. If the modification
573 time has changed, but the size has not, only then does Mercurial need
574 to read the actual contents of the file to see if they've changed.
575 Storing these few extra pieces of information dramatically reduces the
576 amount of data that Mercurial needs to read, which yields large
577 performance improvements compared to other revision control systems.
579 %%% Local Variables:
580 %%% mode: latex
581 %%% TeX-master: "00book"
582 %%% End: