###################################################################### Config::Patcher::Util 0.09 ###################################################################### NAME Config::Patch - Patch configuration files and unpatch them later SYNOPSIS my $patcher = Config::Patch->new( file => "/etc/sudoers", ); # Add a line at the end of /etc/sudoers and reset the file # to its original state later on. my $patch = Config::Patch::Hunk->new( key => "myapp", mode => "append", text => "joeschmoe ALL= NOPASSWD:/etc/rc.d/init.d/myapp", ); $patcher->apply( $patch ); $patcher->save(); # later on: Get /etc/sudoers back to its original state $patcher->eject( "myapp" ); $patcher->save(); DESCRIPTION Config::Patch provides an interface to modify configuration files in a way so that the changes can be rolled back later on. For example, let's say that an application wants to append the line joeschmoe ALL= NOPASSWD:/etc/rc.d/init.d/myapp at the end of the /etc/sudoers file to allow user joeschmoe to start and stop the myapp application as root without having to type a password. Normally, you'd have to do this by hand or via an installation script. And later on, when myapp gets ejected from the system, you'd have to remember to delete the line from /etc/sudoers as well. Config::Patch provides an automated way to apply this 'patch' to the configuration file and to detect and eject it later on. It does this by placing special markers around the insertion, without having to refer to any external meta data. Note that a 'patch' in this context is just a snippet of text that's to be applied somewhere within the file (not to be confused with the diff-formatted text used by the "patch" Unix utility). Patches are line-based, "Config::Patch" always adds/removes entire lines. Command line usage To facilitate its usage, "Config::Patch" comes with a command line script that performs all functions: # Append a patch echo "my patch text" | config-patch -a -k key -f textfile # Patch a file by search-and-replace echo "none:" | config-patch -s 'all:.*' -k key -f config_file # Comment out sections matched by a regular expression: config-patch -c '(?ms-xi:^all:.*?\n\n)' -k key -f config_file # Remove a previously applied patch config-patch -r -k key -f textfile You can only patch a file *once* with a given key. Note that a single patch might result in multiple patched sections within a file if you're using the "replace()" or "comment_out()" methods. To apply different patches to the same file, use different keys. They can be can rolled back separately. API usage With Config::Patch, you run my $patcher = Config::Patch->new( file => "/etc/sudoers", ); to create a patcher object and then define a patch that appends a line to the end of the file: my $patch = Config::Patch::Hunk->new( key => "myapp", mode => "append", text => "joeschmoe ALL= NOPASSWD:/etc/rc.d/init.d/myapp", ); After applying the patch and saving the changes back to the original file, the patch will be in place: $patcher->apply( $patch ); $patcher->save(); along with markers that allow Config::Patch to identify the patch later on and update or eject it: /etc/sudoers *------------------------------------------------ | ... | previous content | ... | #(Config::Patch-myapp-append) | joeschmoe ALL= NOPASSWD:/etc/rc.d/init.d/myapp | #(Config::Patch-myapp-append) *------------------------------------------------ The markers are commented out by '#' marks, and are hence ignored by the application reading the configuration file. However, Config::Patch uses them later on to identify and expunge the comments from the file. To remove the patch from the file later on, call $patcher->eject( "myapp" ); $patcher->save(); and the patcher will scrub the new patch from /etc/sudoers and reset the file to its original state. The "save()" method will write back the file under the name of the currently active file. The path to this file was either set in the Config::Patch constructor with the "file" parameter, or gets set later explicitly via the "file($path)" accessor. If you want to save patched content under a different name, use $patcher->save_as("newfile.dat"); This will also modify the current file setting, which means that if you use read() or save() later on, it will use the newly set name. To peek at the manipulated output before (or after) it's been written, use "$patcher->data()" which returns the current state of the patcher's text data. Patch hunks can be applied to a file in several ways, as specified in the hunk's "mode" field. "append" adds the patch at the end of the file, "prepend" at the beginning, and "replace" searches for a regular expression within the file and then replaces it by the patch. For details on application modes, see the Config::Patch::Hunk section below. Methods "new()" Creates a new Config::Patch object. Takes an optional "file" parameter to specify the 'current' file Config::Patch operates on. "file()" Accessor for the path to the current file. Supports read and write. "read()" Read the current file into memory. Called automatically by apply() if no data has been read into memory yet. "data()" Return the text data Config::Patch is operating on. "save()" Write back the data to the current file. "save_as( $file )" Write back the data to a file named $file. Sets $file as the current file. "apply( $patch )" Applies a patch (a Config::Patch::Hunk object) to the data. "eject( $key )" Removes a patch applied previously under the specified key $key. Instead of a key string, it optionally takes a Config::Patch::Hunk object. "eject_all()" Remove all patches from the data. "parse()" Returns a list of all applied patches so far as Config::Patch::Hunk objects. for my $patch ( $patcher->parse() ) { print $patch->text(); } "patched( $key )" Checks if a patch with the given key was applied to the data already and returns a true value if so. Config::Patch::Hunk Objects "key()" Returns/sets the key under which the key will be applied. The key serves to identify the hunk and to distinguish it from other hunks when identifying/updating/removing the hunk later. "mode()" How the patch will be applied to the data. Supported modes are "append" Add the patch at the end of the data. "prepend" Insert the patch at the beginning of the data. "replace" Replace line ranges matching the regular expression in "regex" with the patch. Encode the replaced data and store it in the patch header, so that it can be put back into place, when the patch is ejected later. For example, to, replace the 'all:' target in a Makefile and all of its production rules by a dummy rule, use my $hunk = Config::Patch::Hunk->new( key => "myapp", mode => "replace", regex => qr(^all:.*?\n\n)sm), text => "all:\n\techo 'all is gone!'\n", ); $patcher->apply( $hunk ); to transform Makefile (before) *------------------------------------------------ | all: | do-this-and-that *------------------------------------------------ into Makefile (after) *------------------------------------------------ | #(Config::Patch-myapp-replace) | all: | echo 'all is gone!' | #(Config::Patch::replace) | # YWxsOgoJZG8tdGhpcy1hbmQtdGhhdAoK | #(Config::Patch::replace) | #(Config::Patch-myapp-replace) *------------------------------------------------ Note the Base64 encoding which carries the original content of the replace line. To remove the patch, run $patcher->eject( "myapp" ); $patcher->save(); and the original content of Makefile will be restored: Makefile (restored) *------------------------------------------------ | all: | do-this-and-that *------------------------------------------------ Tip: To have a hunk comment out a section of the data without adding anything to replace it, simply use an empty "text" field in "replace" mode. "insert-after" Inserts the hunk after a line matching the regular expression defined in "rregex". # Insert "foo=bar" into "[section]". my $hunk = Config::Patch::Hunk->new( key => "myapp", mode => "insert-after", regex => qr(^\[section\])m, text => "foo=bar", ); $patcher->apply( $hunk ); transforms [section] blah into [section] #(Config::Patch-myapp-insert) foo=bar #(Config::Patch-myapp-insert) blah "insert-before" Inserts the hunk *before* a line matching the regular expression defined in $regex. # Insert a new section before [section] my $hunk = Config::Patch::Hunk->new( key => "myapp", mode => "insert-before", regex => qr(^\[section\])m, text => "[newsection]\nfoo=bar\n\n" ); $patcher->apply( $hunk ); transforms [section] blah into #(Config::Patch-myapp-insert) [newsection] foo=bar #(Config::Patch-myapp-insert) [section] blah blah "update" Finds existing hunks and updates them with new values. # Update "myapp" hunk with new value my $hunk = Config::Patch::Hunk->new( key => "myapp", mode => "update", text => "foo=woot", ); While this could be done by removing the hunk via "eject" and then adding it, "update" makes sure the hunk stays exactly in place. "regex" Patch locations are all lines (or line ranges for multi-line regexes) matching the regular expression in "regex" (qr/.../). "text()" The content text the hunk adds to the data. Stripping everything but the hunks The method "$patcher->patches_only()" will trim the surrounding text from the data and just leave the patched sections in place. Patches in Memory Config::Patcher isn't limited to operating on files, you can just as well operate solely in memory. The "data()" method is a read/write accessor to the data string the patcher works on. my $patcher = Config::Patch->new(); $patcher->data( "line1\n", "line2\n" ); $patcher->apply( $patch ); print $patcher->data(); Updating patches Applying a patch if a patch with the same key has already been applied results in an error. For this purpose, use a hunk with the mode field set to 'update'. Newline issues Config::Patch operates line-based, which requires that every line ends with a newline. If you read in a file with trailing characters that aren't ended with a newline, Config::Patch will add a newline at the end. The same applies for patches. Patch lines need to be terminated by a newline, if you forget to specify them that way, Config::Patch will correct it for you. Examining patches To find out what hunks have been applied to the data, use the "parse()" method which returns a list of hunks: for my $hunk ( $patcher->parse() ) { print "Found hunk: ", $hunk->text(), "\n"; } Even after applying a hunk, you have access to a number of updated fields: print "Hunk inserted between positions ", $hunk->pos_start(), " and ", $hunk->pos_end(), "\n"; Both "pos_start" and "pos_end" refer to offsets *including* the markers Config::Patch applies around the content. To find the location of the content of the patch, use "pos_content_start" and "pos_content_end" instead. To obtain the entire text of the hunk (including patch headers), use "as_string()". Specify a different comment character "Config::Patch" assumes that lines starting with a comment character are ignored by their applications. This is important, since "Config::Patch" uses comment lines to hides vital patch information in them for recovering from patches later on. By default, this comment character is '#', usable for file formats like YAML, Makefiles, and Perl. To change this default and use a different character, specify the comment character like my $patcher = Config::Patch->new( comment_char => ";", # comment char is now ";" # ... ); in the constructor call. The command line script "config-patch" expects a different comment character with the "-C" option, check its manpage for details. Make sure to use the same comment character for patching and unpatching, otherwise chaos will ensue. Other than that, "Config::Patch" is format-agnostic. If you need to pay attention to the syntax of the configuration file to be patched, create a subclass of "Config::Patch" and put the format specific logic there. COPYRIGHT AND LICENSE Copyright 2005-2010 by Mike Schilli. This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself. AUTHOR 2005, Mike Schilli