1.1 --- a/es/concepts.tex	Sat Oct 18 07:48:21 2008 -0500
     1.2 +++ b/es/concepts.tex	Sun Oct 19 22:39:43 2008 -0500
     1.3 @@ -0,0 +1,577 @@
     1.4 +\chapter{Behind the scenes}
     1.5 +\label{chap:concepts}
     1.6 +
     1.7 +Unlike many revision control systems, the concepts upon which
     1.8 +Mercurial is built are simple enough that it's easy to understand how
     1.9 +the software really works.  Knowing this certainly isn't necessary,
    1.10 +but I find it useful to have a ``mental model'' of what's going on.
    1.11 +
    1.12 +This understanding gives me confidence that Mercurial has been
    1.13 +carefully designed to be both \emph{safe} and \emph{efficient}.  And
    1.14 +just as importantly, if it's easy for me to retain a good idea of what
    1.15 +the software is doing when I perform a revision control task, I'm less
    1.16 +likely to be surprised by its behaviour.
    1.17 +
    1.18 +In this chapter, we'll initially cover the core concepts behind
    1.19 +Mercurial's design, then continue to discuss some of the interesting
    1.20 +details of its implementation.
    1.21 +
    1.22 +\section{Mercurial's historical record}
    1.23 +
    1.24 +\subsection{Tracking the history of a single file}
    1.25 +
    1.26 +When Mercurial tracks modifications to a file, it stores the history
    1.27 +of that file in a metadata object called a \emph{filelog}.  Each entry
    1.28 +in the filelog contains enough information to reconstruct one revision
    1.29 +of the file that is being tracked.  Filelogs are stored as files in
    1.30 +the \sdirname{.hg/store/data} directory.  A filelog contains two kinds
    1.31 +of information: revision data, and an index to help Mercurial to find
    1.32 +a revision efficiently.
    1.33 +
    1.34 +A file that is large, or has a lot of history, has its filelog stored
    1.35 +in separate data (``\texttt{.d}'' suffix) and index (``\texttt{.i}''
    1.36 +suffix) files.  For small files without much history, the revision
    1.37 +data and index are combined in a single ``\texttt{.i}'' file.  The
    1.38 +correspondence between a file in the working directory and the filelog
    1.39 +that tracks its history in the repository is illustrated in
    1.40 +figure~\ref{fig:concepts:filelog}.
    1.41 +
    1.42 +\begin{figure}[ht]
    1.43 +  \centering
    1.44 +  \grafix{filelog}
    1.45 +  \caption{Relationships between files in working directory and
    1.46 +    filelogs in repository}
    1.47 +  \label{fig:concepts:filelog}
    1.48 +\end{figure}
    1.49 +
    1.50 +\subsection{Managing tracked files}
    1.51 +
    1.52 +Mercurial uses a structure called a \emph{manifest} to collect
    1.53 +together information about the files that it tracks.  Each entry in
    1.54 +the manifest contains information about the files present in a single
    1.55 +changeset.  An entry records which files are present in the changeset,
    1.56 +the revision of each file, and a few other pieces of file metadata.
    1.57 +
    1.58 +\subsection{Recording changeset information}
    1.59 +
    1.60 +The \emph{changelog} contains information about each changeset.  Each
    1.61 +revision records who committed a change, the changeset comment, other
    1.62 +pieces of changeset-related information, and the revision of the
    1.63 +manifest to use.
    1.64 +
    1.65 +\subsection{Relationships between revisions}
    1.66 +
    1.67 +Within a changelog, a manifest, or a filelog, each revision stores a
    1.68 +pointer to its immediate parent (or to its two parents, if it's a
    1.69 +merge revision).  As I mentioned above, there are also relationships
    1.70 +between revisions \emph{across} these structures, and they are
    1.71 +hierarchical in nature.
    1.72 +
    1.73 +For every changeset in a repository, there is exactly one revision
    1.74 +stored in the changelog.  Each revision of the changelog contains a
    1.75 +pointer to a single revision of the manifest.  A revision of the
    1.76 +manifest stores a pointer to a single revision of each filelog tracked
    1.77 +when that changeset was created.  These relationships are illustrated
    1.78 +in figure~\ref{fig:concepts:metadata}.
    1.79 +
    1.80 +\begin{figure}[ht]
    1.81 +  \centering
    1.82 +  \grafix{metadata}
    1.83 +  \caption{Metadata relationships}
    1.84 +  \label{fig:concepts:metadata}
    1.85 +\end{figure}
    1.86 +
    1.87 +As the illustration shows, there is \emph{not} a ``one to one''
    1.88 +relationship between revisions in the changelog, manifest, or filelog.
    1.89 +If the manifest hasn't changed between two changesets, the changelog
    1.90 +entries for those changesets will point to the same revision of the
    1.91 +manifest.  If a file that Mercurial tracks hasn't changed between two
    1.92 +changesets, the entry for that file in the two revisions of the
    1.93 +manifest will point to the same revision of its filelog.
    1.94 +
    1.95 +\section{Safe, efficient storage}
    1.96 +
    1.97 +The underpinnings of changelogs, manifests, and filelogs are provided
    1.98 +by a single structure called the \emph{revlog}.
    1.99 +
   1.100 +\subsection{Efficient storage}
   1.101 +
   1.102 +The revlog provides efficient storage of revisions using a
   1.103 +\emph{delta} mechanism.  Instead of storing a complete copy of a file
   1.104 +for each revision, it stores the changes needed to transform an older
   1.105 +revision into the new revision.  For many kinds of file data, these
   1.106 +deltas are typically a fraction of a percent of the size of a full
   1.107 +copy of a file.
   1.108 +
   1.109 +Some obsolete revision control systems can only work with deltas of
   1.110 +text files.  They must either store binary files as complete snapshots
   1.111 +or encoded into a text representation, both of which are wasteful
   1.112 +approaches.  Mercurial can efficiently handle deltas of files with
   1.113 +arbitrary binary contents; it doesn't need to treat text as special.
   1.114 +
   1.115 +\subsection{Safe operation}
   1.116 +\label{sec:concepts:txn}
   1.117 +
   1.118 +Mercurial only ever \emph{appends} data to the end of a revlog file.
   1.119 +It never modifies a section of a file after it has written it.  This
   1.120 +is both more robust and efficient than schemes that need to modify or
   1.121 +rewrite data.
   1.122 +
   1.123 +In addition, Mercurial treats every write as part of a
   1.124 +\emph{transaction} that can span a number of files.  A transaction is
   1.125 +\emph{atomic}: either the entire transaction succeeds and its effects
   1.126 +are all visible to readers in one go, or the whole thing is undone.
   1.127 +This guarantee of atomicity means that if you're running two copies of
   1.128 +Mercurial, where one is reading data and one is writing it, the reader
   1.129 +will never see a partially written result that might confuse it.
   1.130 +
   1.131 +The fact that Mercurial only appends to files makes it easier to
   1.132 +provide this transactional guarantee.  The easier it is to do stuff
   1.133 +like this, the more confident you should be that it's done correctly.
   1.134 +
   1.135 +\subsection{Fast retrieval}
   1.136 +
   1.137 +Mercurial cleverly avoids a pitfall common to all earlier
   1.138 +revision control systems: the problem of \emph{inefficient retrieval}.
   1.139 +Most revision control systems store the contents of a revision as an
   1.140 +incremental series of modifications against a ``snapshot''.  To
   1.141 +reconstruct a specific revision, you must first read the snapshot, and
   1.142 +then every one of the revisions between the snapshot and your target
   1.143 +revision.  The more history that a file accumulates, the more
   1.144 +revisions you must read, hence the longer it takes to reconstruct a
   1.145 +particular revision.
   1.146 +
   1.147 +\begin{figure}[ht]
   1.148 +  \centering
   1.149 +  \grafix{snapshot}
   1.150 +  \caption{Snapshot of a revlog, with incremental deltas}
   1.151 +  \label{fig:concepts:snapshot}
   1.152 +\end{figure}
   1.153 +
   1.154 +The innovation that Mercurial applies to this problem is simple but
   1.155 +effective.  Once the cumulative amount of delta information stored
   1.156 +since the last snapshot exceeds a fixed threshold, it stores a new
   1.157 +snapshot (compressed, of course), instead of another delta.  This
   1.158 +makes it possible to reconstruct \emph{any} revision of a file
   1.159 +quickly.  This approach works so well that it has since been copied by
   1.160 +several other revision control systems.
   1.161 +
   1.162 +Figure~\ref{fig:concepts:snapshot} illustrates the idea.  In an entry
   1.163 +in a revlog's index file, Mercurial stores the range of entries from
   1.164 +the data file that it must read to reconstruct a particular revision.
   1.165 +
   1.166 +\subsubsection{Aside: the influence of video compression}
   1.167 +
   1.168 +If you're familiar with video compression or have ever watched a TV
   1.169 +feed through a digital cable or satellite service, you may know that
   1.170 +most video compression schemes store each frame of video as a delta
   1.171 +against its predecessor frame.  In addition, these schemes use
   1.172 +``lossy'' compression techniques to increase the compression ratio, so
   1.173 +visual errors accumulate over the course of a number of inter-frame
   1.174 +deltas.
   1.175 +
   1.176 +Because it's possible for a video stream to ``drop out'' occasionally
   1.177 +due to signal glitches, and to limit the accumulation of artefacts
   1.178 +introduced by the lossy compression process, video encoders
   1.179 +periodically insert a complete frame (called a ``key frame'') into the
   1.180 +video stream; the next delta is generated against that frame.  This
   1.181 +means that if the video signal gets interrupted, it will resume once
   1.182 +the next key frame is received.  Also, the accumulation of encoding
   1.183 +errors restarts anew with each key frame.
   1.184 +
   1.185 +\subsection{Identification and strong integrity}
   1.186 +
   1.187 +Along with delta or snapshot information, a revlog entry contains a
   1.188 +cryptographic hash of the data that it represents.  This makes it
   1.189 +difficult to forge the contents of a revision, and easy to detect
   1.190 +accidental corruption.  
   1.191 +
   1.192 +Hashes provide more than a mere check against corruption; they are
   1.193 +used as the identifiers for revisions.  The changeset identification
   1.194 +hashes that you see as an end user are from revisions of the
   1.195 +changelog.  Although filelogs and the manifest also use hashes,
   1.196 +Mercurial only uses these behind the scenes.
   1.197 +
   1.198 +Mercurial verifies that hashes are correct when it retrieves file
   1.199 +revisions and when it pulls changes from another repository.  If it
   1.200 +encounters an integrity problem, it will complain and stop whatever
   1.201 +it's doing.
   1.202 +
   1.203 +In addition to the effect it has on retrieval efficiency, Mercurial's
   1.204 +use of periodic snapshots makes it more robust against partial data
   1.205 +corruption.  If a revlog becomes partly corrupted due to a hardware
   1.206 +error or system bug, it's often possible to reconstruct some or most
   1.207 +revisions from the uncorrupted sections of the revlog, both before and
   1.208 +after the corrupted section.  This would not be possible with a
   1.209 +delta-only storage model.
   1.210 +
   1.211 +\section{Revision history, branching,
   1.212 +  and merging}
   1.213 +
   1.214 +Every entry in a Mercurial revlog knows the identity of its immediate
   1.215 +ancestor revision, usually referred to as its \emph{parent}.  In fact,
   1.216 +a revision contains room for not one parent, but two.  Mercurial uses
   1.217 +a special hash, called the ``null ID'', to represent the idea ``there
   1.218 +is no parent here''.  This hash is simply a string of zeroes.
   1.219 +
   1.220 +In figure~\ref{fig:concepts:revlog}, you can see an example of the
   1.221 +conceptual structure of a revlog.  Filelogs, manifests, and changelogs
   1.222 +all have this same structure; they differ only in the kind of data
   1.223 +stored in each delta or snapshot.
   1.224 +
   1.225 +The first revision in a revlog (at the bottom of the image) has the
   1.226 +null ID in both of its parent slots.  For a ``normal'' revision, its
   1.227 +first parent slot contains the ID of its parent revision, and its
   1.228 +second contains the null ID, indicating that the revision has only one
   1.229 +real parent.  Any two revisions that have the same parent ID are
   1.230 +branches.  A revision that represents a merge between branches has two
   1.231 +normal revision IDs in its parent slots.
   1.232 +
   1.233 +\begin{figure}[ht]
   1.234 +  \centering
   1.235 +  \grafix{revlog}
   1.236 +  \caption{}
   1.237 +  \label{fig:concepts:revlog}
   1.238 +\end{figure}
   1.239 +
   1.240 +\section{The working directory}
   1.241 +
   1.242 +In the working directory, Mercurial stores a snapshot of the files
   1.243 +from the repository as of a particular changeset.
   1.244 +
   1.245 +The working directory ``knows'' which changeset it contains.  When you
   1.246 +update the working directory to contain a particular changeset,
   1.247 +Mercurial looks up the appropriate revision of the manifest to find
   1.248 +out which files it was tracking at the time that changeset was
   1.249 +committed, and which revision of each file was then current.  It then
   1.250 +recreates a copy of each of those files, with the same contents it had
   1.251 +when the changeset was committed.
   1.252 +
   1.253 +The \emph{dirstate} contains Mercurial's knowledge of the working
   1.254 +directory.  This details which changeset the working directory is
   1.255 +updated to, and all of the files that Mercurial is tracking in the
   1.256 +working directory.
   1.257 +
   1.258 +Just as a revision of a revlog has room for two parents, so that it
   1.259 +can represent either a normal revision (with one parent) or a merge of
   1.260 +two earlier revisions, the dirstate has slots for two parents.  When
   1.261 +you use the \hgcmd{update} command, the changeset that you update to
   1.262 +is stored in the ``first parent'' slot, and the null ID in the second.
   1.263 +When you \hgcmd{merge} with another changeset, the first parent
   1.264 +remains unchanged, and the second parent is filled in with the
   1.265 +changeset you're merging with.  The \hgcmd{parents} command tells you
   1.266 +what the parents of the dirstate are.
   1.267 +
   1.268 +\subsection{What happens when you commit}
   1.269 +
   1.270 +The dirstate stores parent information for more than just book-keeping
   1.271 +purposes.  Mercurial uses the parents of the dirstate as \emph{the
   1.272 +  parents of a new changeset} when you perform a commit.
   1.273 +
   1.274 +\begin{figure}[ht]
   1.275 +  \centering
   1.276 +  \grafix{wdir}
   1.277 +  \caption{The working directory can have two parents}
   1.278 +  \label{fig:concepts:wdir}
   1.279 +\end{figure}
   1.280 +
   1.281 +Figure~\ref{fig:concepts:wdir} shows the normal state of the working
   1.282 +directory, where it has a single changeset as parent.  That changeset
   1.283 +is the \emph{tip}, the newest changeset in the repository that has no
   1.284 +children.
   1.285 +
   1.286 +\begin{figure}[ht]
   1.287 +  \centering
   1.288 +  \grafix{wdir-after-commit}
   1.289 +  \caption{The working directory gains new parents after a commit}
   1.290 +  \label{fig:concepts:wdir-after-commit}
   1.291 +\end{figure}
   1.292 +
   1.293 +It's useful to think of the working directory as ``the changeset I'm
   1.294 +about to commit''.  Any files that you tell Mercurial that you've
   1.295 +added, removed, renamed, or copied will be reflected in that
   1.296 +changeset, as will modifications to any files that Mercurial is
   1.297 +already tracking; the new changeset will have the parents of the
   1.298 +working directory as its parents.
   1.299 +
   1.300 +After a commit, Mercurial will update the parents of the working
   1.301 +directory, so that the first parent is the ID of the new changeset,
   1.302 +and the second is the null ID.  This is shown in
   1.303 +figure~\ref{fig:concepts:wdir-after-commit}.  Mercurial doesn't touch
   1.304 +any of the files in the working directory when you commit; it just
   1.305 +modifies the dirstate to note its new parents.
   1.306 +
   1.307 +\subsection{Creating a new head}
   1.308 +
   1.309 +It's perfectly normal to update the working directory to a changeset
   1.310 +other than the current tip.  For example, you might want to know what
   1.311 +your project looked like last Tuesday, or you could be looking through
   1.312 +changesets to see which one introduced a bug.  In cases like this, the
   1.313 +natural thing to do is update the working directory to the changeset
   1.314 +you're interested in, and then examine the files in the working
   1.315 +directory directly to see their contents as they werea when you
   1.316 +committed that changeset.  The effect of this is shown in
   1.317 +figure~\ref{fig:concepts:wdir-pre-branch}.
   1.318 +
   1.319 +\begin{figure}[ht]
   1.320 +  \centering
   1.321 +  \grafix{wdir-pre-branch}
   1.322 +  \caption{The working directory, updated to an older changeset}
   1.323 +  \label{fig:concepts:wdir-pre-branch}
   1.324 +\end{figure}
   1.325 +
   1.326 +Having updated the working directory to an older changeset, what
   1.327 +happens if you make some changes, and then commit?  Mercurial behaves
   1.328 +in the same way as I outlined above.  The parents of the working
   1.329 +directory become the parents of the new changeset.  This new changeset
   1.330 +has no children, so it becomes the new tip.  And the repository now
   1.331 +contains two changesets that have no children; we call these
   1.332 +\emph{heads}.  You can see the structure that this creates in
   1.333 +figure~\ref{fig:concepts:wdir-branch}.
   1.334 +
   1.335 +\begin{figure}[ht]
   1.336 +  \centering
   1.337 +  \grafix{wdir-branch}
   1.338 +  \caption{After a commit made while synced to an older changeset}
   1.339 +  \label{fig:concepts:wdir-branch}
   1.340 +\end{figure}
   1.341 +
   1.342 +\begin{note}
   1.343 +  If you're new to Mercurial, you should keep in mind a common
   1.344 +  ``error'', which is to use the \hgcmd{pull} command without any
   1.345 +  options.  By default, the \hgcmd{pull} command \emph{does not}
   1.346 +  update the working directory, so you'll bring new changesets into
   1.347 +  your repository, but the working directory will stay synced at the
   1.348 +  same changeset as before the pull.  If you make some changes and
   1.349 +  commit afterwards, you'll thus create a new head, because your
   1.350 +  working directory isn't synced to whatever the current tip is.
   1.351 +
   1.352 +  I put the word ``error'' in quotes because all that you need to do
   1.353 +  to rectify this situation is \hgcmd{merge}, then \hgcmd{commit}.  In
   1.354 +  other words, this almost never has negative consequences; it just
   1.355 +  surprises people.  I'll discuss other ways to avoid this behaviour,
   1.356 +  and why Mercurial behaves in this initially surprising way, later
   1.357 +  on.
   1.358 +\end{note}
   1.359 +
   1.360 +\subsection{Merging heads}
   1.361 +
   1.362 +When you run the \hgcmd{merge} command, Mercurial leaves the first
   1.363 +parent of the working directory unchanged, and sets the second parent
   1.364 +to the changeset you're merging with, as shown in
   1.365 +figure~\ref{fig:concepts:wdir-merge}.
   1.366 +
   1.367 +\begin{figure}[ht]
   1.368 +  \centering
   1.369 +  \grafix{wdir-merge}
   1.370 +  \caption{Merging two heads}
   1.371 +  \label{fig:concepts:wdir-merge}
   1.372 +\end{figure}
   1.373 +
   1.374 +Mercurial also has to modify the working directory, to merge the files
   1.375 +managed in the two changesets.  Simplified a little, the merging
   1.376 +process goes like this, for every file in the manifests of both
   1.377 +changesets.
   1.378 +\begin{itemize}
   1.379 +\item If neither changeset has modified a file, do nothing with that
   1.380 +  file.
   1.381 +\item If one changeset has modified a file, and the other hasn't,
   1.382 +  create the modified copy of the file in the working directory.
   1.383 +\item If one changeset has removed a file, and the other hasn't (or
   1.384 +  has also deleted it), delete the file from the working directory.
   1.385 +\item If one changeset has removed a file, but the other has modified
   1.386 +  the file, ask the user what to do: keep the modified file, or remove
   1.387 +  it?
   1.388 +\item If both changesets have modified a file, invoke an external
   1.389 +  merge program to choose the new contents for the merged file.  This
   1.390 +  may require input from the user.
   1.391 +\item If one changeset has modified a file, and the other has renamed
   1.392 +  or copied the file, make sure that the changes follow the new name
   1.393 +  of the file.
   1.394 +\end{itemize}
   1.395 +There are more details---merging has plenty of corner cases---but
   1.396 +these are the most common choices that are involved in a merge.  As
   1.397 +you can see, most cases are completely automatic, and indeed most
   1.398 +merges finish automatically, without requiring your input to resolve
   1.399 +any conflicts.
   1.400 +
   1.401 +When you're thinking about what happens when you commit after a merge,
   1.402 +once again the working directory is ``the changeset I'm about to
   1.403 +commit''.  After the \hgcmd{merge} command completes, the working
   1.404 +directory has two parents; these will become the parents of the new
   1.405 +changeset.
   1.406 +
   1.407 +Mercurial lets you perform multiple merges, but you must commit the
   1.408 +results of each individual merge as you go.  This is necessary because
   1.409 +Mercurial only tracks two parents for both revisions and the working
   1.410 +directory.  While it would be technically possible to merge multiple
   1.411 +changesets at once, the prospect of user confusion and making a
   1.412 +terrible mess of a merge immediately becomes overwhelming.
   1.413 +
   1.414 +\section{Other interesting design features}
   1.415 +
   1.416 +In the sections above, I've tried to highlight some of the most
   1.417 +important aspects of Mercurial's design, to illustrate that it pays
   1.418 +careful attention to reliability and performance.  However, the
   1.419 +attention to detail doesn't stop there.  There are a number of other
   1.420 +aspects of Mercurial's construction that I personally find
   1.421 +interesting.  I'll detail a few of them here, separate from the ``big
   1.422 +ticket'' items above, so that if you're interested, you can gain a
   1.423 +better idea of the amount of thinking that goes into a well-designed
   1.424 +system.
   1.425 +
   1.426 +\subsection{Clever compression}
   1.427 +
   1.428 +When appropriate, Mercurial will store both snapshots and deltas in
   1.429 +compressed form.  It does this by always \emph{trying to} compress a
   1.430 +snapshot or delta, but only storing the compressed version if it's
   1.431 +smaller than the uncompressed version.
   1.432 +
   1.433 +This means that Mercurial does ``the right thing'' when storing a file
   1.434 +whose native form is compressed, such as a \texttt{zip} archive or a
   1.435 +JPEG image.  When these types of files are compressed a second time,
   1.436 +the resulting file is usually bigger than the once-compressed form,
   1.437 +and so Mercurial will store the plain \texttt{zip} or JPEG.
   1.438 +
   1.439 +Deltas between revisions of a compressed file are usually larger than
   1.440 +snapshots of the file, and Mercurial again does ``the right thing'' in
   1.441 +these cases.  It finds that such a delta exceeds the threshold at
   1.442 +which it should store a complete snapshot of the file, so it stores
   1.443 +the snapshot, again saving space compared to a naive delta-only
   1.444 +approach.
   1.445 +
   1.446 +\subsubsection{Network recompression}
   1.447 +
   1.448 +When storing revisions on disk, Mercurial uses the ``deflate''
   1.449 +compression algorithm (the same one used by the popular \texttt{zip}
   1.450 +archive format), which balances good speed with a respectable
   1.451 +compression ratio.  However, when transmitting revision data over a
   1.452 +network connection, Mercurial uncompresses the compressed revision
   1.453 +data.
   1.454 +
   1.455 +If the connection is over HTTP, Mercurial recompresses the entire
   1.456 +stream of data using a compression algorithm that gives a better
   1.457 +compression ratio (the Burrows-Wheeler algorithm from the widely used
   1.458 +\texttt{bzip2} compression package).  This combination of algorithm
   1.459 +and compression of the entire stream (instead of a revision at a time)
   1.460 +substantially reduces the number of bytes to be transferred, yielding
   1.461 +better network performance over almost all kinds of network.
   1.462 +
   1.463 +(If the connection is over \command{ssh}, Mercurial \emph{doesn't}
   1.464 +recompress the stream, because \command{ssh} can already do this
   1.465 +itself.)
   1.466 +
   1.467 +\subsection{Read/write ordering and atomicity}
   1.468 +
   1.469 +Appending to files isn't the whole story when it comes to guaranteeing
   1.470 +that a reader won't see a partial write.  If you recall
   1.471 +figure~\ref{fig:concepts:metadata}, revisions in the changelog point to
   1.472 +revisions in the manifest, and revisions in the manifest point to
   1.473 +revisions in filelogs.  This hierarchy is deliberate.
   1.474 +
   1.475 +A writer starts a transaction by writing filelog and manifest data,
   1.476 +and doesn't write any changelog data until those are finished.  A
   1.477 +reader starts by reading changelog data, then manifest data, followed
   1.478 +by filelog data.
   1.479 +
   1.480 +Since the writer has always finished writing filelog and manifest data
   1.481 +before it writes to the changelog, a reader will never read a pointer
   1.482 +to a partially written manifest revision from the changelog, and it will
   1.483 +never read a pointer to a partially written filelog revision from the
   1.484 +manifest.
   1.485 +
   1.486 +\subsection{Concurrent access}
   1.487 +
   1.488 +The read/write ordering and atomicity guarantees mean that Mercurial
   1.489 +never needs to \emph{lock} a repository when it's reading data, even
   1.490 +if the repository is being written to while the read is occurring.
   1.491 +This has a big effect on scalability; you can have an arbitrary number
   1.492 +of Mercurial processes safely reading data from a repository safely
   1.493 +all at once, no matter whether it's being written to or not.
   1.494 +
   1.495 +The lockless nature of reading means that if you're sharing a
   1.496 +repository on a multi-user system, you don't need to grant other local
   1.497 +users permission to \emph{write} to your repository in order for them
   1.498 +to be able to clone it or pull changes from it; they only need
   1.499 +\emph{read} permission.  (This is \emph{not} a common feature among
   1.500 +revision control systems, so don't take it for granted!  Most require
   1.501 +readers to be able to lock a repository to access it safely, and this
   1.502 +requires write permission on at least one directory, which of course
   1.503 +makes for all kinds of nasty and annoying security and administrative
   1.504 +problems.)
   1.505 +
   1.506 +Mercurial uses locks to ensure that only one process can write to a
   1.507 +repository at a time (the locking mechanism is safe even over
   1.508 +filesystems that are notoriously hostile to locking, such as NFS).  If
   1.509 +a repository is locked, a writer will wait for a while to retry if the
   1.510 +repository becomes unlocked, but if the repository remains locked for
   1.511 +too long, the process attempting to write will time out after a while.
   1.512 +This means that your daily automated scripts won't get stuck forever
   1.513 +and pile up if a system crashes unnoticed, for example.  (Yes, the
   1.514 +timeout is configurable, from zero to infinity.)
   1.515 +
   1.516 +\subsubsection{Safe dirstate access}
   1.517 +
   1.518 +As with revision data, Mercurial doesn't take a lock to read the
   1.519 +dirstate file; it does acquire a lock to write it.  To avoid the
   1.520 +possibility of reading a partially written copy of the dirstate file,
   1.521 +Mercurial writes to a file with a unique name in the same directory as
   1.522 +the dirstate file, then renames the temporary file atomically to
   1.523 +\filename{dirstate}.  The file named \filename{dirstate} is thus
   1.524 +guaranteed to be complete, not partially written.
   1.525 +
   1.526 +\subsection{Avoiding seeks}
   1.527 +
   1.528 +Critical to Mercurial's performance is the avoidance of seeks of the
   1.529 +disk head, since any seek is far more expensive than even a
   1.530 +comparatively large read operation.
   1.531 +
   1.532 +This is why, for example, the dirstate is stored in a single file.  If
   1.533 +there were a dirstate file per directory that Mercurial tracked, the
   1.534 +disk would seek once per directory.  Instead, Mercurial reads the
   1.535 +entire single dirstate file in one step.
   1.536 +
   1.537 +Mercurial also uses a ``copy on write'' scheme when cloning a
   1.538 +repository on local storage.  Instead of copying every revlog file
   1.539 +from the old repository into the new repository, it makes a ``hard
   1.540 +link'', which is a shorthand way to say ``these two names point to the
   1.541 +same file''.  When Mercurial is about to write to one of a revlog's
   1.542 +files, it checks to see if the number of names pointing at the file is
   1.543 +greater than one.  If it is, more than one repository is using the
   1.544 +file, so Mercurial makes a new copy of the file that is private to
   1.545 +this repository.
   1.546 +
   1.547 +A few revision control developers have pointed out that this idea of
   1.548 +making a complete private copy of a file is not very efficient in its
   1.549 +use of storage.  While this is true, storage is cheap, and this method
   1.550 +gives the highest performance while deferring most book-keeping to the
   1.551 +operating system.  An alternative scheme would most likely reduce
   1.552 +performance and increase the complexity of the software, each of which
   1.553 +is much more important to the ``feel'' of day-to-day use.
   1.554 +
   1.555 +\subsection{Other contents of the dirstate}
   1.556 +
   1.557 +Because Mercurial doesn't force you to tell it when you're modifying a
   1.558 +file, it uses the dirstate to store some extra information so it can
   1.559 +determine efficiently whether you have modified a file.  For each file
   1.560 +in the working directory, it stores the time that it last modified the
   1.561 +file itself, and the size of the file at that time.  
   1.562 +
   1.563 +When you explicitly \hgcmd{add}, \hgcmd{remove}, \hgcmd{rename} or
   1.564 +\hgcmd{copy} files, Mercurial updates the dirstate so that it knows
   1.565 +what to do with those files when you commit.
   1.566 +
   1.567 +When Mercurial is checking the states of files in the working
   1.568 +directory, it first checks a file's modification time.  If that has
   1.569 +not changed, the file must not have been modified.  If the file's size
   1.570 +has changed, the file must have been modified.  If the modification
   1.571 +time has changed, but the size has not, only then does Mercurial need
   1.572 +to read the actual contents of the file to see if they've changed.
   1.573 +Storing these few extra pieces of information dramatically reduces the
   1.574 +amount of data that Mercurial needs to read, which yields large
   1.575 +performance improvements compared to other revision control systems.
   1.576 +
   1.577 +%%% Local Variables: 
   1.578 +%%% mode: latex
   1.579 +%%% TeX-master: "00book"
   1.580 +%%% End: