merge original branch.

[tortoisegit/TortoiseGitJp.git] / doc / source / en / TortoiseGit / git_doc / git-fast-import.html.xml

<?xml version="1.0" encoding="UTF-8"?>\r
<!DOCTYPE article PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN" "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd">\r
\r
<article lang="en" id="git-fast-import(1)">\r
<articleinfo>\r
    <title>git-fast-import(1)</title>\r
        <indexterm>\r
                <primary>git-fast-import(1)</primary>\r
        </indexterm>\r
</articleinfo>\r
<simplesect id="_name">\r
<title>NAME</title>\r
<simpara>git-fast-import - Backend for fast Git data importers</simpara>\r
</simplesect>\r
<simplesect id="_synopsis">\r
<title>SYNOPSIS</title>\r
<simpara>frontend | <emphasis>git fast-import</emphasis> [options]</simpara>\r
</simplesect>\r
<simplesect id="_description">\r
<title>DESCRIPTION</title>\r
<simpara>This program is usually not what the end user wants to run directly.\r
Most end users want to use one of the existing frontend programs,\r
which parses a specific type of foreign source and feeds the contents\r
stored there to <emphasis>git-fast-import</emphasis>.</simpara>\r
<simpara>fast-import reads a mixed command/data stream from standard input and\r
writes one or more packfiles directly into the current repository.\r
When EOF is received on standard input, fast import writes out\r
updated branch and tag refs, fully updating the current repository\r
with the newly imported data.</simpara>\r
<simpara>The fast-import backend itself can import into an empty repository (one that\r
has already been initialized by <emphasis>git-init</emphasis>) or incrementally\r
update an existing populated repository.  Whether or not incremental\r
imports are supported from a particular foreign source depends on\r
the frontend program in use.</simpara>\r
</simplesect>\r
<simplesect id="_options">\r
<title>OPTIONS</title>\r
<variablelist>\r
<varlistentry>\r
<term>\r
--date-format=&lt;fmt&gt;\r
</term>\r
<listitem>\r
<simpara>\r
        Specify the type of dates the frontend will supply to\r
        fast-import within <literal>author</literal>, <literal>committer</literal> and <literal>tagger</literal> commands.\r
        See &#8220;Date Formats&#8221; below for details about which formats\r
        are supported, and their syntax.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
--force\r
</term>\r
<listitem>\r
<simpara>\r
        Force updating modified existing branches, even if doing\r
        so would cause commits to be lost (as the new commit does\r
        not contain the old commit).\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
--max-pack-size=&lt;n&gt;\r
</term>\r
<listitem>\r
<simpara>\r
        Maximum size of each output packfile, expressed in MiB.\r
        The default is 4096 (4 GiB) as that is the maximum allowed\r
        packfile size (due to file format limitations). Some\r
        importers may wish to lower this, such as to ensure the\r
        resulting packfiles fit on CDs.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
--depth=&lt;n&gt;\r
</term>\r
<listitem>\r
<simpara>\r
        Maximum delta depth, for blob and tree deltification.\r
        Default is 10.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
--active-branches=&lt;n&gt;\r
</term>\r
<listitem>\r
<simpara>\r
        Maximum number of branches to maintain active at once.\r
        See &#8220;Memory Utilization&#8221; below for details.  Default is 5.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
--export-marks=&lt;file&gt;\r
</term>\r
<listitem>\r
<simpara>\r
        Dumps the internal marks table to &lt;file&gt; when complete.\r
        Marks are written one per line as <literal>:markid SHA-1</literal>.\r
        Frontends can use this file to validate imports after they\r
        have been completed, or to save the marks table across\r
        incremental runs.  As &lt;file&gt; is only opened and truncated\r
        at checkpoint (or completion) the same path can also be\r
        safely given to --import-marks.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
--import-marks=&lt;file&gt;\r
</term>\r
<listitem>\r
<simpara>\r
        Before processing any input, load the marks specified in\r
        &lt;file&gt;.  The input file must exist, must be readable, and\r
        must use the same format as produced by --export-marks.\r
        Multiple options may be supplied to import more than one\r
        set of marks.  If a mark is defined to different values,\r
        the last file wins.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
--export-pack-edges=&lt;file&gt;\r
</term>\r
<listitem>\r
<simpara>\r
        After creating a packfile, print a line of data to\r
        &lt;file&gt; listing the filename of the packfile and the last\r
        commit on each branch that was written to that packfile.\r
        This information may be useful after importing projects\r
        whose total object set exceeds the 4 GiB packfile limit,\r
        as these commits can be used as edge points during calls\r
        to <emphasis>git-pack-objects</emphasis>.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
--quiet\r
</term>\r
<listitem>\r
<simpara>\r
        Disable all non-fatal output, making fast-import silent when it\r
        is successful.  This option disables the output shown by\r
        --stats.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
--stats\r
</term>\r
<listitem>\r
<simpara>\r
        Display some basic statistics about the objects fast-import has\r
        created, the packfiles they were stored into, and the\r
        memory used by fast-import during this run.  Showing this output\r
        is currently the default, but can be disabled with --quiet.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
</variablelist>\r
</simplesect>\r
<simplesect id="_performance">\r
<title>Performance</title>\r
<simpara>The design of fast-import allows it to import large projects in a minimum\r
amount of memory usage and processing time.  Assuming the frontend\r
is able to keep up with fast-import and feed it a constant stream of data,\r
import times for projects holding 10+ years of history and containing\r
100,000+ individual commits are generally completed in just 1-2\r
hours on quite modest (~$2,000 USD) hardware.</simpara>\r
<simpara>Most bottlenecks appear to be in foreign source data access (the\r
source just cannot extract revisions fast enough) or disk IO (fast-import\r
writes as fast as the disk will take the data).  Imports will run\r
faster if the source data is stored on a different drive than the\r
destination Git repository (due to less IO contention).</simpara>\r
</simplesect>\r
<simplesect id="_development_cost">\r
<title>Development Cost</title>\r
<simpara>A typical frontend for fast-import tends to weigh in at approximately 200\r
lines of Perl/Python/Ruby code.  Most developers have been able to\r
create working importers in just a couple of hours, even though it\r
is their first exposure to fast-import, and sometimes even to Git.  This is\r
an ideal situation, given that most conversion tools are throw-away\r
(use once, and never look back).</simpara>\r
</simplesect>\r
<simplesect id="_parallel_operation">\r
<title>Parallel Operation</title>\r
<simpara>Like <emphasis>git-push</emphasis> or <emphasis>git-fetch</emphasis>, imports handled by fast-import are safe to\r
run alongside parallel <literal>git repack -a -d</literal> or <literal>git gc</literal> invocations,\r
or any other Git operation (including <emphasis>git-prune</emphasis>, as loose objects\r
are never used by fast-import).</simpara>\r
<simpara>fast-import does not lock the branch or tag refs it is actively importing.\r
After the import, during its ref update phase, fast-import tests each\r
existing branch ref to verify the update will be a fast-forward\r
update (the commit stored in the ref is contained in the new\r
history of the commit to be written).  If the update is not a\r
fast-forward update, fast-import will skip updating that ref and instead\r
prints a warning message.  fast-import will always attempt to update all\r
branch refs, and does not stop on the first failure.</simpara>\r
<simpara>Branch updates can be forced with --force, but its recommended that\r
this only be used on an otherwise quiet repository.  Using --force\r
is not necessary for an initial import into an empty repository.</simpara>\r
</simplesect>\r
<simplesect id="_technical_discussion">\r
<title>Technical Discussion</title>\r
<simpara>fast-import tracks a set of branches in memory.  Any branch can be created\r
or modified at any point during the import process by sending a\r
<literal>commit</literal> command on the input stream.  This design allows a frontend\r
program to process an unlimited number of branches simultaneously,\r
generating commits in the order they are available from the source\r
data.  It also simplifies the frontend programs considerably.</simpara>\r
<simpara>fast-import does not use or alter the current working directory, or any\r
file within it.  (It does however update the current Git repository,\r
as referenced by <literal>GIT_DIR</literal>.)  Therefore an import frontend may use\r
the working directory for its own purposes, such as extracting file\r
revisions from the foreign source.  This ignorance of the working\r
directory also allows fast-import to run very quickly, as it does not\r
need to perform any costly file update operations when switching\r
between branches.</simpara>\r
</simplesect>\r
<simplesect id="_input_format">\r
<title>Input Format</title>\r
<simpara>With the exception of raw file data (which Git does not interpret)\r
the fast-import input format is text (ASCII) based.  This text based\r
format simplifies development and debugging of frontend programs,\r
especially when a higher level language such as Perl, Python or\r
Ruby is being used.</simpara>\r
<simpara>fast-import is very strict about its input.  Where we say SP below we mean\r
<emphasis role="strong">exactly</emphasis> one space.  Likewise LF means one (and only one) linefeed.\r
Supplying additional whitespace characters will cause unexpected\r
results, such as branch names or file names with leading or trailing\r
spaces in their name, or early termination of fast-import when it encounters\r
unexpected input.</simpara>\r
<simplesect id="_stream_comments">\r
<title>Stream Comments</title>\r
<simpara>To aid in debugging frontends fast-import ignores any line that\r
begins with <literal>#</literal> (ASCII pound/hash) up to and including the line\r
ending <literal>LF</literal>.  A comment line may contain any sequence of bytes\r
that does not contain an LF and therefore may be used to include\r
any detailed debugging information that might be specific to the\r
frontend and useful when inspecting a fast-import data stream.</simpara>\r
</simplesect>\r
<simplesect id="_date_formats">\r
<title>Date Formats</title>\r
<simpara>The following date formats are supported.  A frontend should select\r
the format it will use for this import by passing the format name\r
in the --date-format=&lt;fmt&gt; command line option.</simpara>\r
<variablelist>\r
<varlistentry>\r
<term>\r
<literal>raw</literal>\r
</term>\r
<listitem>\r
<simpara>\r
        This is the Git native format and is <literal>&lt;time&gt; SP &lt;offutc&gt;</literal>.\r
        It is also fast-import&#8217;s default format, if --date-format was\r
        not specified.\r
</simpara>\r
<simpara>The time of the event is specified by <literal>&lt;time&gt;</literal> as the number of\r
seconds since the UNIX epoch (midnight, Jan 1, 1970, UTC) and is\r
written as an ASCII decimal integer.</simpara>\r
<simpara>The local offset is specified by <literal>&lt;offutc&gt;</literal> as a positive or negative\r
offset from UTC.  For example EST (which is 5 hours behind UTC)\r
would be expressed in <literal>&lt;tz&gt;</literal> by &#8220;-0500&#8221; while UTC is &#8220;+0000&#8221;.\r
The local offset does not affect <literal>&lt;time&gt;</literal>; it is used only as an\r
advisement to help formatting routines display the timestamp.</simpara>\r
<simpara>If the local offset is not available in the source material, use\r
&#8220;+0000&#8221;, or the most common local offset.  For example many\r
organizations have a CVS repository which has only ever been accessed\r
by users who are located in the same location and timezone.  In this\r
case a reasonable offset from UTC could be assumed.</simpara>\r
<simpara>Unlike the <literal>rfc2822</literal> format, this format is very strict.  Any\r
variation in formatting will cause fast-import to reject the value.</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
<literal>rfc2822</literal>\r
</term>\r
<listitem>\r
<simpara>\r
        This is the standard email format as described by RFC 2822.\r
</simpara>\r
<simpara>An example value is &#8220;Tue Feb 6 11:22:18 2007 -0500&#8221;.  The Git\r
parser is accurate, but a little on the lenient side.  It is the\r
same parser used by <emphasis>git-am</emphasis> when applying patches\r
received from email.</simpara>\r
<simpara>Some malformed strings may be accepted as valid dates.  In some of\r
these cases Git will still be able to obtain the correct date from\r
the malformed string.  There are also some types of malformed\r
strings which Git will parse wrong, and yet consider valid.\r
Seriously malformed strings will be rejected.</simpara>\r
<simpara>Unlike the <literal>raw</literal> format above, the timezone/UTC offset information\r
contained in an RFC 2822 date string is used to adjust the date\r
value to UTC prior to storage.  Therefore it is important that\r
this information be as accurate as possible.</simpara>\r
<simpara>If the source material uses RFC 2822 style dates,\r
the frontend should let fast-import handle the parsing and conversion\r
(rather than attempting to do it itself) as the Git parser has\r
been well tested in the wild.</simpara>\r
<simpara>Frontends should prefer the <literal>raw</literal> format if the source material\r
already uses UNIX-epoch format, can be coaxed to give dates in that\r
format, or its format is easily convertible to it, as there is no\r
ambiguity in parsing.</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
<literal>now</literal>\r
</term>\r
<listitem>\r
<simpara>\r
        Always use the current time and timezone.  The literal\r
        <literal>now</literal> must always be supplied for <literal>&lt;when&gt;</literal>.\r
</simpara>\r
<simpara>This is a toy format.  The current time and timezone of this system\r
is always copied into the identity string at the time it is being\r
created by fast-import.  There is no way to specify a different time or\r
timezone.</simpara>\r
<simpara>This particular format is supplied as its short to implement and\r
may be useful to a process that wants to create a new commit\r
right now, without needing to use a working directory or\r
<emphasis>git-update-index</emphasis>.</simpara>\r
<simpara>If separate <literal>author</literal> and <literal>committer</literal> commands are used in a <literal>commit</literal>\r
the timestamps may not match, as the system clock will be polled\r
twice (once for each command).  The only way to ensure that both\r
author and committer identity information has the same timestamp\r
is to omit <literal>author</literal> (thus copying from <literal>committer</literal>) or to use a\r
date format other than <literal>now</literal>.</simpara>\r
</listitem>\r
</varlistentry>\r
</variablelist>\r
</simplesect>\r
<simplesect id="_commands">\r
<title>Commands</title>\r
<simpara>fast-import accepts several commands to update the current repository\r
and control the current import process.  More detailed discussion\r
(with examples) of each command follows later.</simpara>\r
<variablelist>\r
<varlistentry>\r
<term>\r
<literal>commit</literal>\r
</term>\r
<listitem>\r
<simpara>\r
        Creates a new branch or updates an existing branch by\r
        creating a new commit and updating the branch to point at\r
        the newly created commit.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
<literal>tag</literal>\r
</term>\r
<listitem>\r
<simpara>\r
        Creates an annotated tag object from an existing commit or\r
        branch.  Lightweight tags are not supported by this command,\r
        as they are not recommended for recording meaningful points\r
        in time.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
<literal>reset</literal>\r
</term>\r
<listitem>\r
<simpara>\r
        Reset an existing branch (or a new branch) to a specific\r
        revision.  This command must be used to change a branch to\r
        a specific revision without making a commit on it.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
<literal>blob</literal>\r
</term>\r
<listitem>\r
<simpara>\r
        Convert raw file data into a blob, for future use in a\r
        <literal>commit</literal> command.  This command is optional and is not\r
        needed to perform an import.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
<literal>checkpoint</literal>\r
</term>\r
<listitem>\r
<simpara>\r
        Forces fast-import to close the current packfile, generate its\r
        unique SHA-1 checksum and index, and start a new packfile.\r
        This command is optional and is not needed to perform\r
        an import.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
<literal>progress</literal>\r
</term>\r
<listitem>\r
<simpara>\r
        Causes fast-import to echo the entire line to its own\r
        standard output.  This command is optional and is not needed\r
        to perform an import.\r
</simpara>\r
</listitem>\r
</varlistentry>\r
</variablelist>\r
</simplesect>\r
<simplesect id="_literal_commit_literal">\r
<title><literal>commit</literal></title>\r
<simpara>Create or update a branch with a new commit, recording one logical\r
change to the project.</simpara>\r
<literallayout class="monospaced">        'commit' SP &lt;ref&gt; LF\r
        mark?\r
        ('author' SP &lt;name&gt; SP LT &lt;email&gt; GT SP &lt;when&gt; LF)?\r
        'committer' SP &lt;name&gt; SP LT &lt;email&gt; GT SP &lt;when&gt; LF\r
        data\r
        ('from' SP &lt;committish&gt; LF)?\r
        ('merge' SP &lt;committish&gt; LF)?\r
        (filemodify | filedelete | filecopy | filerename | filedeleteall)*\r
        LF?</literallayout>\r
<simpara>where <literal>&lt;ref&gt;</literal> is the name of the branch to make the commit on.\r
Typically branch names are prefixed with <literal>refs/heads/</literal> in\r
Git, so importing the CVS branch symbol <literal>RELENG-1_0</literal> would use\r
<literal>refs/heads/RELENG-1_0</literal> for the value of <literal>&lt;ref&gt;</literal>.  The value of\r
<literal>&lt;ref&gt;</literal> must be a valid refname in Git.  As <literal>LF</literal> is not valid in\r
a Git refname, no quoting or escaping syntax is supported here.</simpara>\r
<simpara>A <literal>mark</literal> command may optionally appear, requesting fast-import to save a\r
reference to the newly created commit for future use by the frontend\r
(see below for format).  It is very common for frontends to mark\r
every commit they create, thereby allowing future branch creation\r
from any imported commit.</simpara>\r
<simpara>The <literal>data</literal> command following <literal>committer</literal> must supply the commit\r
message (see below for <literal>data</literal> command syntax).  To import an empty\r
commit message use a 0 length data.  Commit messages are free-form\r
and are not interpreted by Git.  Currently they must be encoded in\r
UTF-8, as fast-import does not permit other encodings to be specified.</simpara>\r
<simpara>Zero or more <literal>filemodify</literal>, <literal>filedelete</literal>, <literal>filecopy</literal>, <literal>filerename</literal>\r
and <literal>filedeleteall</literal> commands\r
may be included to update the contents of the branch prior to\r
creating the commit.  These commands may be supplied in any order.\r
However it is recommended that a <literal>filedeleteall</literal> command precede\r
all <literal>filemodify</literal>, <literal>filecopy</literal> and <literal>filerename</literal> commands in the same\r
commit, as <literal>filedeleteall</literal>\r
wipes the branch clean (see below).</simpara>\r
<simpara>The <literal>LF</literal> after the command is optional (it used to be required).</simpara>\r
<simplesect id="_literal_author_literal">\r
<title><literal>author</literal></title>\r
<simpara>An <literal>author</literal> command may optionally appear, if the author information\r
might differ from the committer information.  If <literal>author</literal> is omitted\r
then fast-import will automatically use the committer&#8217;s information for\r
the author portion of the commit.  See below for a description of\r
the fields in <literal>author</literal>, as they are identical to <literal>committer</literal>.</simpara>\r
</simplesect>\r
<simplesect id="_literal_committer_literal">\r
<title><literal>committer</literal></title>\r
<simpara>The <literal>committer</literal> command indicates who made this commit, and when\r
they made it.</simpara>\r
<simpara>Here <literal>&lt;name&gt;</literal> is the person&#8217;s display name (for example\r
&#8220;Com M Itter&#8221;) and <literal>&lt;email&gt;</literal> is the person&#8217;s email address\r
(&#8220;<ulink url="mailto:cm@example.com">cm@example.com</ulink>&#8221;).  <literal>LT</literal> and <literal>GT</literal> are the literal less-than (\x3c)\r
and greater-than (\x3e) symbols.  These are required to delimit\r
the email address from the other fields in the line.  Note that\r
<literal>&lt;name&gt;</literal> is free-form and may contain any sequence of bytes, except\r
<literal>LT</literal> and <literal>LF</literal>.  It is typically UTF-8 encoded.</simpara>\r
<simpara>The time of the change is specified by <literal>&lt;when&gt;</literal> using the date format\r
that was selected by the --date-format=&lt;fmt&gt; command line option.\r
See &#8220;Date Formats&#8221; above for the set of supported formats, and\r
their syntax.</simpara>\r
</simplesect>\r
<simplesect id="_literal_from_literal">\r
<title><literal>from</literal></title>\r
<simpara>The <literal>from</literal> command is used to specify the commit to initialize\r
this branch from.  This revision will be the first ancestor of the\r
new commit.</simpara>\r
<simpara>Omitting the <literal>from</literal> command in the first commit of a new branch\r
will cause fast-import to create that commit with no ancestor. This\r
tends to be desired only for the initial commit of a project.\r
If the frontend creates all files from scratch when making a new\r
branch, a <literal>merge</literal> command may be used instead of <literal>from</literal> to start\r
the commit with an empty tree.\r
Omitting the <literal>from</literal> command on existing branches is usually desired,\r
as the current commit on that branch is automatically assumed to\r
be the first ancestor of the new commit.</simpara>\r
<simpara>As <literal>LF</literal> is not valid in a Git refname or SHA-1 expression, no\r
quoting or escaping syntax is supported within <literal>&lt;committish&gt;</literal>.</simpara>\r
<simpara>Here <literal>&lt;committish&gt;</literal> is any of the following:</simpara>\r
<itemizedlist>\r
<listitem>\r
<simpara>\r
The name of an existing branch already in fast-import&#8217;s internal branch\r
  table.  If fast-import doesn&#8217;t know the name, its treated as a SHA-1\r
  expression.\r
</simpara>\r
</listitem>\r
<listitem>\r
<simpara>\r
A mark reference, <literal>:&lt;idnum&gt;</literal>, where <literal>&lt;idnum&gt;</literal> is the mark number.\r
</simpara>\r
<simpara>The reason fast-import uses <literal>:</literal> to denote a mark reference is this character\r
is not legal in a Git branch name.  The leading <literal>:</literal> makes it easy\r
to distinguish between the mark 42 (<literal>:42</literal>) and the branch 42 (<literal>42</literal>\r
or <literal>refs/heads/42</literal>), or an abbreviated SHA-1 which happened to\r
consist only of base-10 digits.</simpara>\r
<simpara>Marks must be declared (via <literal>mark</literal>) before they can be used.</simpara>\r
</listitem>\r
<listitem>\r
<simpara>\r
A complete 40 byte or abbreviated commit SHA-1 in hex.\r
</simpara>\r
</listitem>\r
<listitem>\r
<simpara>\r
Any valid Git SHA-1 expression that resolves to a commit.  See\r
  &#8220;SPECIFYING REVISIONS&#8221; in <xref linkend="git-rev-parse(1)"/> for details.\r
</simpara>\r
</listitem>\r
</itemizedlist>\r
<simpara>The special case of restarting an incremental import from the\r
current branch value should be written as:</simpara>\r
<literallayout>        from refs/heads/branch^0</literallayout>\r
<simpara>The <literal>&#94;0</literal> suffix is necessary as fast-import does not permit a branch to\r
start from itself, and the branch is created in memory before the\r
<literal>from</literal> command is even read from the input.  Adding <literal>&#94;0</literal> will force\r
fast-import to resolve the commit through Git&#8217;s revision parsing library,\r
rather than its internal branch table, thereby loading in the\r
existing value of the branch.</simpara>\r
</simplesect>\r
<simplesect id="_literal_merge_literal">\r
<title><literal>merge</literal></title>\r
<simpara>Includes one additional ancestor commit.  If the <literal>from</literal> command is\r
omitted when creating a new branch, the first <literal>merge</literal> commit will be\r
the first ancestor of the current commit, and the branch will start\r
out with no files.  An unlimited number of <literal>merge</literal> commands per\r
commit are permitted by fast-import, thereby establishing an n-way merge.\r
However Git&#8217;s other tools never create commits with more than 15\r
additional ancestors (forming a 16-way merge).  For this reason\r
it is suggested that frontends do not use more than 15 <literal>merge</literal>\r
commands per commit; 16, if starting a new, empty branch.</simpara>\r
<simpara>Here <literal>&lt;committish&gt;</literal> is any of the commit specification expressions\r
also accepted by <literal>from</literal> (see above).</simpara>\r
</simplesect>\r
<simplesect id="_literal_filemodify_literal">\r
<title><literal>filemodify</literal></title>\r
<simpara>Included in a <literal>commit</literal> command to add a new file or change the\r
content of an existing file.  This command has two different means\r
of specifying the content of the file.</simpara>\r
<variablelist>\r
<varlistentry>\r
<term>\r
External data format\r
</term>\r
<listitem>\r
<simpara>\r
        The data content for the file was already supplied by a prior\r
        <literal>blob</literal> command.  The frontend just needs to connect it.\r
</simpara>\r
<literallayout class="monospaced">        'M' SP &lt;mode&gt; SP &lt;dataref&gt; SP &lt;path&gt; LF</literallayout>\r
<simpara>Here <literal>&lt;dataref&gt;</literal> can be either a mark reference (<literal>:&lt;idnum&gt;</literal>)\r
set by a prior <literal>blob</literal> command, or a full 40-byte SHA-1 of an\r
existing Git blob object.</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
Inline data format\r
</term>\r
<listitem>\r
<simpara>\r
        The data content for the file has not been supplied yet.\r
        The frontend wants to supply it as part of this modify\r
        command.\r
</simpara>\r
<literallayout class="monospaced">        'M' SP &lt;mode&gt; SP 'inline' SP &lt;path&gt; LF\r
        data</literallayout>\r
<simpara>See below for a detailed description of the <literal>data</literal> command.</simpara>\r
</listitem>\r
</varlistentry>\r
</variablelist>\r
<simpara>In both formats <literal>&lt;mode&gt;</literal> is the type of file entry, specified\r
in octal.  Git only supports the following modes:</simpara>\r
<itemizedlist>\r
<listitem>\r
<simpara>\r
<literal>100644</literal> or <literal>644</literal>: A normal (not-executable) file.  The majority\r
  of files in most projects use this mode.  If in doubt, this is\r
  what you want.\r
</simpara>\r
</listitem>\r
<listitem>\r
<simpara>\r
<literal>100755</literal> or <literal>755</literal>: A normal, but executable, file.\r
</simpara>\r
</listitem>\r
<listitem>\r
<simpara>\r
<literal>120000</literal>: A symlink, the content of the file will be the link target.\r
</simpara>\r
</listitem>\r
<listitem>\r
<simpara>\r
<literal>160000</literal>: A gitlink, SHA-1 of the object refers to a commit in\r
  another repository. Git links can only be specified by SHA or through\r
  a commit mark. They are used to implement submodules.\r
</simpara>\r
</listitem>\r
</itemizedlist>\r
<simpara>In both formats <literal>&lt;path&gt;</literal> is the complete path of the file to be added\r
(if not already existing) or modified (if already existing).</simpara>\r
<simpara>A <literal>&lt;path&gt;</literal> string must use UNIX-style directory separators (forward\r
slash <literal>/</literal>), may contain any byte other than <literal>LF</literal>, and must not\r
start with double quote (<literal>"</literal>).</simpara>\r
<simpara>If an <literal>LF</literal> or double quote must be encoded into <literal>&lt;path&gt;</literal> shell-style\r
quoting should be used, e.g. <literal>"path/with\n and \" in it"</literal>.</simpara>\r
<simpara>The value of <literal>&lt;path&gt;</literal> must be in canonical form. That is it must not:</simpara>\r
<itemizedlist>\r
<listitem>\r
<simpara>\r
contain an empty directory component (e.g. <literal>foo//bar</literal> is invalid),\r
</simpara>\r
</listitem>\r
<listitem>\r
<simpara>\r
end with a directory separator (e.g. <literal>foo/</literal> is invalid),\r
</simpara>\r
</listitem>\r
<listitem>\r
<simpara>\r
start with a directory separator (e.g. <literal>/foo</literal> is invalid),\r
</simpara>\r
</listitem>\r
<listitem>\r
<simpara>\r
contain the special component <literal>.</literal> or <literal>..</literal> (e.g. <literal>foo/./bar</literal> and\r
  <literal>foo/../bar</literal> are invalid).\r
</simpara>\r
</listitem>\r
</itemizedlist>\r
<simpara>It is recommended that <literal>&lt;path&gt;</literal> always be encoded using UTF-8.</simpara>\r
</simplesect>\r
<simplesect id="_literal_filedelete_literal">\r
<title><literal>filedelete</literal></title>\r
<simpara>Included in a <literal>commit</literal> command to remove a file or recursively\r
delete an entire directory from the branch.  If the file or directory\r
removal makes its parent directory empty, the parent directory will\r
be automatically removed too.  This cascades up the tree until the\r
first non-empty directory or the root is reached.</simpara>\r
<literallayout class="monospaced">        'D' SP &lt;path&gt; LF</literallayout>\r
<simpara>here <literal>&lt;path&gt;</literal> is the complete path of the file or subdirectory to\r
be removed from the branch.\r
See <literal>filemodify</literal> above for a detailed description of <literal>&lt;path&gt;</literal>.</simpara>\r
</simplesect>\r
<simplesect id="_literal_filecopy_literal">\r
<title><literal>filecopy</literal></title>\r
<simpara>Recursively copies an existing file or subdirectory to a different\r
location within the branch.  The existing file or directory must\r
exist.  If the destination exists it will be completely replaced\r
by the content copied from the source.</simpara>\r
<literallayout class="monospaced">        'C' SP &lt;path&gt; SP &lt;path&gt; LF</literallayout>\r
<simpara>here the first <literal>&lt;path&gt;</literal> is the source location and the second\r
<literal>&lt;path&gt;</literal> is the destination.  See <literal>filemodify</literal> above for a detailed\r
description of what <literal>&lt;path&gt;</literal> may look like.  To use a source path\r
that contains SP the path must be quoted.</simpara>\r
<simpara>A <literal>filecopy</literal> command takes effect immediately.  Once the source\r
location has been copied to the destination any future commands\r
applied to the source location will not impact the destination of\r
the copy.</simpara>\r
</simplesect>\r
<simplesect id="_literal_filerename_literal">\r
<title><literal>filerename</literal></title>\r
<simpara>Renames an existing file or subdirectory to a different location\r
within the branch.  The existing file or directory must exist. If\r
the destination exists it will be replaced by the source directory.</simpara>\r
<literallayout class="monospaced">        'R' SP &lt;path&gt; SP &lt;path&gt; LF</literallayout>\r
<simpara>here the first <literal>&lt;path&gt;</literal> is the source location and the second\r
<literal>&lt;path&gt;</literal> is the destination.  See <literal>filemodify</literal> above for a detailed\r
description of what <literal>&lt;path&gt;</literal> may look like.  To use a source path\r
that contains SP the path must be quoted.</simpara>\r
<simpara>A <literal>filerename</literal> command takes effect immediately.  Once the source\r
location has been renamed to the destination any future commands\r
applied to the source location will create new files there and not\r
impact the destination of the rename.</simpara>\r
<simpara>Note that a <literal>filerename</literal> is the same as a <literal>filecopy</literal> followed by a\r
<literal>filedelete</literal> of the source location.  There is a slight performance\r
advantage to using <literal>filerename</literal>, but the advantage is so small\r
that it is never worth trying to convert a delete/add pair in\r
source material into a rename for fast-import.  This <literal>filerename</literal>\r
command is provided just to simplify frontends that already have\r
rename information and don&#8217;t want bother with decomposing it into a\r
<literal>filecopy</literal> followed by a <literal>filedelete</literal>.</simpara>\r
</simplesect>\r
<simplesect id="_literal_filedeleteall_literal">\r
<title><literal>filedeleteall</literal></title>\r
<simpara>Included in a <literal>commit</literal> command to remove all files (and also all\r
directories) from the branch.  This command resets the internal\r
branch structure to have no files in it, allowing the frontend\r
to subsequently add all interesting files from scratch.</simpara>\r
<literallayout class="monospaced">        'deleteall' LF</literallayout>\r
<simpara>This command is extremely useful if the frontend does not know\r
(or does not care to know) what files are currently on the branch,\r
and therefore cannot generate the proper <literal>filedelete</literal> commands to\r
update the content.</simpara>\r
<simpara>Issuing a <literal>filedeleteall</literal> followed by the needed <literal>filemodify</literal>\r
commands to set the correct content will produce the same results\r
as sending only the needed <literal>filemodify</literal> and <literal>filedelete</literal> commands.\r
The <literal>filedeleteall</literal> approach may however require fast-import to use slightly\r
more memory per active branch (less than 1 MiB for even most large\r
projects); so frontends that can easily obtain only the affected\r
paths for a commit are encouraged to do so.</simpara>\r
</simplesect>\r
</simplesect>\r
<simplesect id="_literal_mark_literal">\r
<title><literal>mark</literal></title>\r
<simpara>Arranges for fast-import to save a reference to the current object, allowing\r
the frontend to recall this object at a future point in time, without\r
knowing its SHA-1.  Here the current object is the object creation\r
command the <literal>mark</literal> command appears within.  This can be <literal>commit</literal>,\r
<literal>tag</literal>, and <literal>blob</literal>, but <literal>commit</literal> is the most common usage.</simpara>\r
<literallayout class="monospaced">        'mark' SP ':' &lt;idnum&gt; LF</literallayout>\r
<simpara>where <literal>&lt;idnum&gt;</literal> is the number assigned by the frontend to this mark.\r
The value of <literal>&lt;idnum&gt;</literal> is expressed as an ASCII decimal integer.\r
The value 0 is reserved and cannot be used as\r
a mark.  Only values greater than or equal to 1 may be used as marks.</simpara>\r
<simpara>New marks are created automatically.  Existing marks can be moved\r
to another object simply by reusing the same <literal>&lt;idnum&gt;</literal> in another\r
<literal>mark</literal> command.</simpara>\r
</simplesect>\r
<simplesect id="_literal_tag_literal">\r
<title><literal>tag</literal></title>\r
<simpara>Creates an annotated tag referring to a specific commit.  To create\r
lightweight (non-annotated) tags see the <literal>reset</literal> command below.</simpara>\r
<literallayout class="monospaced">        'tag' SP &lt;name&gt; LF\r
        'from' SP &lt;committish&gt; LF\r
        'tagger' SP &lt;name&gt; SP LT &lt;email&gt; GT SP &lt;when&gt; LF\r
        data</literallayout>\r
<simpara>where <literal>&lt;name&gt;</literal> is the name of the tag to create.</simpara>\r
<simpara>Tag names are automatically prefixed with <literal>refs/tags/</literal> when stored\r
in Git, so importing the CVS branch symbol <literal>RELENG-1_0-FINAL</literal> would\r
use just <literal>RELENG-1_0-FINAL</literal> for <literal>&lt;name&gt;</literal>, and fast-import will write the\r
corresponding ref as <literal>refs/tags/RELENG-1_0-FINAL</literal>.</simpara>\r
<simpara>The value of <literal>&lt;name&gt;</literal> must be a valid refname in Git and therefore\r
may contain forward slashes.  As <literal>LF</literal> is not valid in a Git refname,\r
no quoting or escaping syntax is supported here.</simpara>\r
<simpara>The <literal>from</literal> command is the same as in the <literal>commit</literal> command; see\r
above for details.</simpara>\r
<simpara>The <literal>tagger</literal> command uses the same format as <literal>committer</literal> within\r
<literal>commit</literal>; again see above for details.</simpara>\r
<simpara>The <literal>data</literal> command following <literal>tagger</literal> must supply the annotated tag\r
message (see below for <literal>data</literal> command syntax).  To import an empty\r
tag message use a 0 length data.  Tag messages are free-form and are\r
not interpreted by Git.  Currently they must be encoded in UTF-8,\r
as fast-import does not permit other encodings to be specified.</simpara>\r
<simpara>Signing annotated tags during import from within fast-import is not\r
supported.  Trying to include your own PGP/GPG signature is not\r
recommended, as the frontend does not (easily) have access to the\r
complete set of bytes which normally goes into such a signature.\r
If signing is required, create lightweight tags from within fast-import with\r
<literal>reset</literal>, then create the annotated versions of those tags offline\r
with the standard <emphasis>git-tag</emphasis> process.</simpara>\r
</simplesect>\r
<simplesect id="_literal_reset_literal">\r
<title><literal>reset</literal></title>\r
<simpara>Creates (or recreates) the named branch, optionally starting from\r
a specific revision.  The reset command allows a frontend to issue\r
a new <literal>from</literal> command for an existing branch, or to create a new\r
branch from an existing commit without creating a new commit.</simpara>\r
<literallayout class="monospaced">        'reset' SP &lt;ref&gt; LF\r
        ('from' SP &lt;committish&gt; LF)?\r
        LF?</literallayout>\r
<simpara>For a detailed description of <literal>&lt;ref&gt;</literal> and <literal>&lt;committish&gt;</literal> see above\r
under <literal>commit</literal> and <literal>from</literal>.</simpara>\r
<simpara>The <literal>LF</literal> after the command is optional (it used to be required).</simpara>\r
<simpara>The <literal>reset</literal> command can also be used to create lightweight\r
(non-annotated) tags.  For example:</simpara>\r
<informalexample>\r
<literallayout class="monospaced">reset refs/tags/938\r
from :938</literallayout>\r
</informalexample>\r
<simpara>would create the lightweight tag <literal>refs/tags/938</literal> referring to\r
whatever commit mark <literal>:938</literal> references.</simpara>\r
</simplesect>\r
<simplesect id="_literal_blob_literal">\r
<title><literal>blob</literal></title>\r
<simpara>Requests writing one file revision to the packfile.  The revision\r
is not connected to any commit; this connection must be formed in\r
a subsequent <literal>commit</literal> command by referencing the blob through an\r
assigned mark.</simpara>\r
<literallayout class="monospaced">        'blob' LF\r
        mark?\r
        data</literallayout>\r
<simpara>The mark command is optional here as some frontends have chosen\r
to generate the Git SHA-1 for the blob on their own, and feed that\r
directly to <literal>commit</literal>.  This is typically more work than its worth\r
however, as marks are inexpensive to store and easy to use.</simpara>\r
</simplesect>\r
<simplesect id="_literal_data_literal">\r
<title><literal>data</literal></title>\r
<simpara>Supplies raw data (for use as blob/file content, commit messages, or\r
annotated tag messages) to fast-import.  Data can be supplied using an exact\r
byte count or delimited with a terminating line.  Real frontends\r
intended for production-quality conversions should always use the\r
exact byte count format, as it is more robust and performs better.\r
The delimited format is intended primarily for testing fast-import.</simpara>\r
<simpara>Comment lines appearing within the <literal>&lt;raw&gt;</literal> part of <literal>data</literal> commands\r
are always taken to be part of the body of the data and are therefore\r
never ignored by fast-import.  This makes it safe to import any\r
file/message content whose lines might start with <literal>#</literal>.</simpara>\r
<variablelist>\r
<varlistentry>\r
<term>\r
Exact byte count format\r
</term>\r
<listitem>\r
<simpara>\r
        The frontend must specify the number of bytes of data.\r
</simpara>\r
<literallayout class="monospaced">        'data' SP &lt;count&gt; LF\r
        &lt;raw&gt; LF?</literallayout>\r
<simpara>where <literal>&lt;count&gt;</literal> is the exact number of bytes appearing within\r
<literal>&lt;raw&gt;</literal>.  The value of <literal>&lt;count&gt;</literal> is expressed as an ASCII decimal\r
integer.  The <literal>LF</literal> on either side of <literal>&lt;raw&gt;</literal> is not\r
included in <literal>&lt;count&gt;</literal> and will not be included in the imported data.</simpara>\r
<simpara>The <literal>LF</literal> after <literal>&lt;raw&gt;</literal> is optional (it used to be required) but\r
recommended.  Always including it makes debugging a fast-import\r
stream easier as the next command always starts in column 0\r
of the next line, even if <literal>&lt;raw&gt;</literal> did not end with an <literal>LF</literal>.</simpara>\r
</listitem>\r
</varlistentry>\r
<varlistentry>\r
<term>\r
Delimited format\r
</term>\r
<listitem>\r
<simpara>\r
        A delimiter string is used to mark the end of the data.\r
        fast-import will compute the length by searching for the delimiter.\r
        This format is primarily useful for testing and is not\r
        recommended for real data.\r
</simpara>\r
<literallayout class="monospaced">        'data' SP '&lt;&lt;' &lt;delim&gt; LF\r
        &lt;raw&gt; LF\r
        &lt;delim&gt; LF\r
        LF?</literallayout>\r
<simpara>where <literal>&lt;delim&gt;</literal> is the chosen delimiter string.  The string <literal>&lt;delim&gt;</literal>\r
must not appear on a line by itself within <literal>&lt;raw&gt;</literal>, as otherwise\r
fast-import will think the data ends earlier than it really does.  The <literal>LF</literal>\r
immediately trailing <literal>&lt;raw&gt;</literal> is part of <literal>&lt;raw&gt;</literal>.  This is one of\r
the limitations of the delimited format, it is impossible to supply\r
a data chunk which does not have an LF as its last byte.</simpara>\r
<simpara>The <literal>LF</literal> after <literal>&lt;delim&gt; LF</literal> is optional (it used to be required).</simpara>\r
</listitem>\r
</varlistentry>\r
</variablelist>\r
</simplesect>\r
<simplesect id="_literal_checkpoint_literal">\r
<title><literal>checkpoint</literal></title>\r
<simpara>Forces fast-import to close the current packfile, start a new one, and to\r
save out all current branch refs, tags and marks.</simpara>\r
<literallayout class="monospaced">        'checkpoint' LF\r
        LF?</literallayout>\r
<simpara>Note that fast-import automatically switches packfiles when the current\r
packfile reaches --max-pack-size, or 4 GiB, whichever limit is\r
smaller.  During an automatic packfile switch fast-import does not update\r
the branch refs, tags or marks.</simpara>\r
<simpara>As a <literal>checkpoint</literal> can require a significant amount of CPU time and\r
disk IO (to compute the overall pack SHA-1 checksum, generate the\r
corresponding index file, and update the refs) it can easily take\r
several minutes for a single <literal>checkpoint</literal> command to complete.</simpara>\r
<simpara>Frontends may choose to issue checkpoints during extremely large\r
and long running imports, or when they need to allow another Git\r
process access to a branch.  However given that a 30 GiB Subversion\r
repository can be loaded into Git through fast-import in about 3 hours,\r
explicit checkpointing may not be necessary.</simpara>\r
<simpara>The <literal>LF</literal> after the command is optional (it used to be required).</simpara>\r
</simplesect>\r
<simplesect id="_literal_progress_literal">\r
<title><literal>progress</literal></title>\r
<simpara>Causes fast-import to print the entire <literal>progress</literal> line unmodified to\r
its standard output channel (file descriptor 1) when the command is\r
processed from the input stream.  The command otherwise has no impact\r
on the current import, or on any of fast-import&#8217;s internal state.</simpara>\r
<literallayout class="monospaced">        'progress' SP &lt;any&gt; LF\r
        LF?</literallayout>\r
<simpara>The <literal>&lt;any&gt;</literal> part of the command may contain any sequence of bytes\r
that does not contain <literal>LF</literal>.  The <literal>LF</literal> after the command is optional.\r
Callers may wish to process the output through a tool such as sed to\r
remove the leading part of the line, for example:</simpara>\r
<informalexample>\r
<literallayout class="monospaced">frontend | git fast-import | sed 's/^progress //'</literallayout>\r
</informalexample>\r
<simpara>Placing a <literal>progress</literal> command immediately after a <literal>checkpoint</literal> will\r
inform the reader when the <literal>checkpoint</literal> has been completed and it\r
can safely access the refs that fast-import updated.</simpara>\r
</simplesect>\r
</simplesect>\r
<simplesect id="_crash_reports">\r
<title>Crash Reports</title>\r
<simpara>If fast-import is supplied invalid input it will terminate with a\r
non-zero exit status and create a crash report in the top level of\r
the Git repository it was importing into.  Crash reports contain\r
a snapshot of the internal fast-import state as well as the most\r
recent commands that lead up to the crash.</simpara>\r
<simpara>All recent commands (including stream comments, file changes and\r
progress commands) are shown in the command history within the crash\r
report, but raw file data and commit messages are excluded from the\r
crash report.  This exclusion saves space within the report file\r
and reduces the amount of buffering that fast-import must perform\r
during execution.</simpara>\r
<simpara>After writing a crash report fast-import will close the current\r
packfile and export the marks table.  This allows the frontend\r
developer to inspect the repository state and resume the import from\r
the point where it crashed.  The modified branches and tags are not\r
updated during a crash, as the import did not complete successfully.\r
Branch and tag information can be found in the crash report and\r
must be applied manually if the update is needed.</simpara>\r
<simpara>An example crash:</simpara>\r
<informalexample>\r
<literallayout class="monospaced">$ cat &gt;in &lt;&lt;END_OF_INPUT\r
# my very first test commit\r
commit refs/heads/master\r
committer Shawn O. Pearce &lt;spearce&gt; 19283 -0400\r
# who is that guy anyway?\r
data &lt;&lt;EOF\r
this is my commit\r
EOF\r
M 644 inline .gitignore\r
data &lt;&lt;EOF\r
.gitignore\r
EOF\r
M 777 inline bob\r
END_OF_INPUT</literallayout>\r
<literallayout class="monospaced">$ git fast-import &lt;in\r
fatal: Corrupt mode: M 777 inline bob\r
fast-import: dumping crash report to .git/fast_import_crash_8434</literallayout>\r
<literallayout class="monospaced">$ cat .git/fast_import_crash_8434\r
fast-import crash report:\r
    fast-import process: 8434\r
    parent process     : 1391\r
    at Sat Sep 1 00:58:12 2007</literallayout>\r
<literallayout class="monospaced">fatal: Corrupt mode: M 777 inline bob</literallayout>\r
<literallayout class="monospaced">Most Recent Commands Before Crash\r
---------------------------------\r
  # my very first test commit\r
  commit refs/heads/master\r
  committer Shawn O. Pearce &lt;spearce&gt; 19283 -0400\r
  # who is that guy anyway?\r
  data &lt;&lt;EOF\r
  M 644 inline .gitignore\r
  data &lt;&lt;EOF\r
* M 777 inline bob</literallayout>\r
<literallayout class="monospaced">Active Branch LRU\r
-----------------\r
    active_branches = 1 cur, 5 max</literallayout>\r
<literallayout class="monospaced">pos  clock name\r
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\r
 1)      0 refs/heads/master</literallayout>\r
<literallayout class="monospaced">Inactive Branches\r
-----------------\r
refs/heads/master:\r
  status      : active loaded dirty\r
  tip commit  : 0000000000000000000000000000000000000000\r
  old tree    : 0000000000000000000000000000000000000000\r
  cur tree    : 0000000000000000000000000000000000000000\r
  commit clock: 0\r
  last pack   :</literallayout>\r
<literallayout class="monospaced">-------------------\r
END OF CRASH REPORT</literallayout>\r
</informalexample>\r
</simplesect>\r
<simplesect id="_tips_and_tricks">\r
<title>Tips and Tricks</title>\r
<simpara>The following tips and tricks have been collected from various\r
users of fast-import, and are offered here as suggestions.</simpara>\r
<simplesect id="_use_one_mark_per_commit">\r
<title>Use One Mark Per Commit</title>\r
<simpara>When doing a repository conversion, use a unique mark per commit\r
(<literal>mark :&lt;n&gt;</literal>) and supply the --export-marks option on the command\r
line.  fast-import will dump a file which lists every mark and the Git\r
object SHA-1 that corresponds to it.  If the frontend can tie\r
the marks back to the source repository, it is easy to verify the\r
accuracy and completeness of the import by comparing each Git\r
commit to the corresponding source revision.</simpara>\r
<simpara>Coming from a system such as Perforce or Subversion this should be\r
quite simple, as the fast-import mark can also be the Perforce changeset\r
number or the Subversion revision number.</simpara>\r
</simplesect>\r
<simplesect id="_freely_skip_around_branches">\r
<title>Freely Skip Around Branches</title>\r
<simpara>Don&#8217;t bother trying to optimize the frontend to stick to one branch\r
at a time during an import.  Although doing so might be slightly\r
faster for fast-import, it tends to increase the complexity of the frontend\r
code considerably.</simpara>\r
<simpara>The branch LRU builtin to fast-import tends to behave very well, and the\r
cost of activating an inactive branch is so low that bouncing around\r
between branches has virtually no impact on import performance.</simpara>\r
</simplesect>\r
<simplesect id="_handling_renames">\r
<title>Handling Renames</title>\r
<simpara>When importing a renamed file or directory, simply delete the old\r
name(s) and modify the new name(s) during the corresponding commit.\r
Git performs rename detection after-the-fact, rather than explicitly\r
during a commit.</simpara>\r
</simplesect>\r
<simplesect id="_use_tag_fixup_branches">\r
<title>Use Tag Fixup Branches</title>\r
<simpara>Some other SCM systems let the user create a tag from multiple\r
files which are not from the same commit/changeset.  Or to create\r
tags which are a subset of the files available in the repository.</simpara>\r
<simpara>Importing these tags as-is in Git is impossible without making at\r
least one commit which &#8220;fixes up&#8221; the files to match the content\r
of the tag.  Use fast-import&#8217;s <literal>reset</literal> command to reset a dummy branch\r
outside of your normal branch space to the base commit for the tag,\r
then commit one or more file fixup commits, and finally tag the\r
dummy branch.</simpara>\r
<simpara>For example since all normal branches are stored under <literal>refs/heads/</literal>\r
name the tag fixup branch <literal>TAG_FIXUP</literal>.  This way it is impossible for\r
the fixup branch used by the importer to have namespace conflicts\r
with real branches imported from the source (the name <literal>TAG_FIXUP</literal>\r
is not <literal>refs/heads/TAG_FIXUP</literal>).</simpara>\r
<simpara>When committing fixups, consider using <literal>merge</literal> to connect the\r
commit(s) which are supplying file revisions to the fixup branch.\r
Doing so will allow tools such as <emphasis>git-blame</emphasis> to track\r
through the real commit history and properly annotate the source\r
files.</simpara>\r
<simpara>After fast-import terminates the frontend will need to do <literal>rm .git/TAG_FIXUP</literal>\r
to remove the dummy branch.</simpara>\r
</simplesect>\r
<simplesect id="_import_now_repack_later">\r
<title>Import Now, Repack Later</title>\r
<simpara>As soon as fast-import completes the Git repository is completely valid\r
and ready for use.  Typically this takes only a very short time,\r
even for considerably large projects (100,000+ commits).</simpara>\r
<simpara>However repacking the repository is necessary to improve data\r
locality and access performance.  It can also take hours on extremely\r
large projects (especially if -f and a large --window parameter is\r
used).  Since repacking is safe to run alongside readers and writers,\r
run the repack in the background and let it finish when it finishes.\r
There is no reason to wait to explore your new Git project!</simpara>\r
<simpara>If you choose to wait for the repack, don&#8217;t try to run benchmarks\r
or performance tests until repacking is completed.  fast-import outputs\r
suboptimal packfiles that are simply never seen in real use\r
situations.</simpara>\r
</simplesect>\r
<simplesect id="_repacking_historical_data">\r
<title>Repacking Historical Data</title>\r
<simpara>If you are repacking very old imported data (e.g. older than the\r
last year), consider expending some extra CPU time and supplying\r
--window=50 (or higher) when you run <emphasis>git-repack</emphasis>.\r
This will take longer, but will also produce a smaller packfile.\r
You only need to expend the effort once, and everyone using your\r
project will benefit from the smaller repository.</simpara>\r
</simplesect>\r
<simplesect id="_include_some_progress_messages">\r
<title>Include Some Progress Messages</title>\r
<simpara>Every once in a while have your frontend emit a <literal>progress</literal> message\r
to fast-import.  The contents of the messages are entirely free-form,\r
so one suggestion would be to output the current month and year\r
each time the current commit date moves into the next month.\r
Your users will feel better knowing how much of the data stream\r
has been processed.</simpara>\r
</simplesect>\r
</simplesect>\r
<simplesect id="_packfile_optimization">\r
<title>Packfile Optimization</title>\r
<simpara>When packing a blob fast-import always attempts to deltify against the last\r
blob written.  Unless specifically arranged for by the frontend,\r
this will probably not be a prior version of the same file, so the\r
generated delta will not be the smallest possible.  The resulting\r
packfile will be compressed, but will not be optimal.</simpara>\r
<simpara>Frontends which have efficient access to all revisions of a\r
single file (for example reading an RCS/CVS ,v file) can choose\r
to supply all revisions of that file as a sequence of consecutive\r
<literal>blob</literal> commands.  This allows fast-import to deltify the different file\r
revisions against each other, saving space in the final packfile.\r
Marks can be used to later identify individual file revisions during\r
a sequence of <literal>commit</literal> commands.</simpara>\r
<simpara>The packfile(s) created by fast-import do not encourage good disk access\r
patterns.  This is caused by fast-import writing the data in the order\r
it is received on standard input, while Git typically organizes\r
data within packfiles to make the most recent (current tip) data\r
appear before historical data.  Git also clusters commits together,\r
speeding up revision traversal through better cache locality.</simpara>\r
<simpara>For this reason it is strongly recommended that users repack the\r
repository with <literal>git repack -a -d</literal> after fast-import completes, allowing\r
Git to reorganize the packfiles for faster data access.  If blob\r
deltas are suboptimal (see above) then also adding the <literal>-f</literal> option\r
to force recomputation of all deltas can significantly reduce the\r
final packfile size (30-50% smaller can be quite typical).</simpara>\r
</simplesect>\r
<simplesect id="_memory_utilization">\r
<title>Memory Utilization</title>\r
<simpara>There are a number of factors which affect how much memory fast-import\r
requires to perform an import.  Like critical sections of core\r
Git, fast-import uses its own memory allocators to amortize any overheads\r
associated with malloc.  In practice fast-import tends to amortize any\r
malloc overheads to 0, due to its use of large block allocations.</simpara>\r
<simplesect id="_per_object">\r
<title>per object</title>\r
<simpara>fast-import maintains an in-memory structure for every object written in\r
this execution.  On a 32 bit system the structure is 32 bytes,\r
on a 64 bit system the structure is 40 bytes (due to the larger\r
pointer sizes).  Objects in the table are not deallocated until\r
fast-import terminates.  Importing 2 million objects on a 32 bit system\r
will require approximately 64 MiB of memory.</simpara>\r
<simpara>The object table is actually a hashtable keyed on the object name\r
(the unique SHA-1).  This storage configuration allows fast-import to reuse\r
an existing or already written object and avoid writing duplicates\r
to the output packfile.  Duplicate blobs are surprisingly common\r
in an import, typically due to branch merges in the source.</simpara>\r
</simplesect>\r
<simplesect id="_per_mark">\r
<title>per mark</title>\r
<simpara>Marks are stored in a sparse array, using 1 pointer (4 bytes or 8\r
bytes, depending on pointer size) per mark.  Although the array\r
is sparse, frontends are still strongly encouraged to use marks\r
between 1 and n, where n is the total number of marks required for\r
this import.</simpara>\r
</simplesect>\r
<simplesect id="_per_branch">\r
<title>per branch</title>\r
<simpara>Branches are classified as active and inactive.  The memory usage\r
of the two classes is significantly different.</simpara>\r
<simpara>Inactive branches are stored in a structure which uses 96 or 120\r
bytes (32 bit or 64 bit systems, respectively), plus the length of\r
the branch name (typically under 200 bytes), per branch.  fast-import will\r
easily handle as many as 10,000 inactive branches in under 2 MiB\r
of memory.</simpara>\r
<simpara>Active branches have the same overhead as inactive branches, but\r
also contain copies of every tree that has been recently modified on\r
that branch.  If subtree <literal>include</literal> has not been modified since the\r
branch became active, its contents will not be loaded into memory,\r
but if subtree <literal>src</literal> has been modified by a commit since the branch\r
became active, then its contents will be loaded in memory.</simpara>\r
<simpara>As active branches store metadata about the files contained on that\r
branch, their in-memory storage size can grow to a considerable size\r
(see below).</simpara>\r
<simpara>fast-import automatically moves active branches to inactive status based on\r
a simple least-recently-used algorithm.  The LRU chain is updated on\r
each <literal>commit</literal> command.  The maximum number of active branches can be\r
increased or decreased on the command line with --active-branches=.</simpara>\r
</simplesect>\r
<simplesect id="_per_active_tree">\r
<title>per active tree</title>\r
<simpara>Trees (aka directories) use just 12 bytes of memory on top of the\r
memory required for their entries (see &#8220;per active file&#8221; below).\r
The cost of a tree is virtually 0, as its overhead amortizes out\r
over the individual file entries.</simpara>\r
</simplesect>\r
<simplesect id="_per_active_file_entry">\r
<title>per active file entry</title>\r
<simpara>Files (and pointers to subtrees) within active trees require 52 or 64\r
bytes (32/64 bit platforms) per entry.  To conserve space, file and\r
tree names are pooled in a common string table, allowing the filename\r
&#8220;Makefile&#8221; to use just 16 bytes (after including the string header\r
overhead) no matter how many times it occurs within the project.</simpara>\r
<simpara>The active branch LRU, when coupled with the filename string pool\r
and lazy loading of subtrees, allows fast-import to efficiently import\r
projects with 2,000+ branches and 45,114+ files in a very limited\r
memory footprint (less than 2.7 MiB per active branch).</simpara>\r
</simplesect>\r
</simplesect>\r
<simplesect id="_author">\r
<title>Author</title>\r
<simpara>Written by Shawn O. Pearce &lt;<ulink url="mailto:spearce@spearce.org">spearce@spearce.org</ulink>&gt;.</simpara>\r
</simplesect>\r
<simplesect id="_documentation">\r
<title>Documentation</title>\r
<simpara>Documentation by Shawn O. Pearce &lt;<ulink url="mailto:spearce@spearce.org">spearce@spearce.org</ulink>&gt;.</simpara>\r
</simplesect>\r
<simplesect id="_git">\r
<title>GIT</title>\r
<simpara>Part of the <xref linkend="git(1)"/> suite</simpara>\r
</simplesect>\r
</article>\r