Fix PHP serialized objects in MySQL dumps after find/replace.

serfix corrects the character counts in PHP serialized string objects, within a MySQL dump file (.sql), when you need to do a mass change of the strings themselves.

serfix was conceived out of a need to automate complex Wordpress development and deployment workflows for a large number of projects. Specifically, when changing environments between development/staging/production, a find/replace tool (such as sed) can be used on the database dumps to update the site's root URL, however this will generally have unforseen and undesireable consequences when it breaks the PHP serialized objects in the dump file.

serfix was originally a Python script, but for a variety of reasons it was desirable to write it in a compiled binary form.

Download the latest compiled release from releases and put it in your PATH (somewhere like /usr/local/bin will usually work on OSX).

-or-

go get github.com/astockwell/serfix

-or-

1. Clone & cd into the repo
2. go install serfix.go (requires go compiler installed)

serfix can be used as a stdin line filter (e.g. chained together with pipes and other unix commands) or as a standalone program.

serfix [flags] filename [outfilename]

• -f, --force: Force overwrite of destination file if it exists (only used with [outfilename] specified)
• -h, --help: Print serfix help

cat filename.sql | serfix > fixed_filename.sql

ssh -C user@host mysqldump --single-transaction --opt \
--net_buffer_length=75000 -u'username' -p'password' db_name \
| sed 's/development.com/production.com/g' | serfix | gzip > \
db_name_$(date +"%Y.%m.%d_%H.%M").sql.gz

serfix myfile.sql   # saves over source file

serfix myfile.sql new_filename.sql   # saves to new file

serfix -f myfile.sql existing_filename.sql   # saves over destination file

Gives you a .sql file full of s:20:"http://mydevsite.com"; PHP serialized string objects.

Gives you a .sql file full of s:20:"http://myproductionsite.com"; objects that PHP will reject because the character count doesn't match the string.

Gives you a .sql file full of s:27:"http://myproductionsite.com"; objects with the character count correctly matching the string.

High fives all around.

serfix uses a 2MB buffer of RAM, so line lengths exceeding 2MB in your .sql file or stdin will cause an error. This can be mitigated by using the --net_buffer_length=75000 flag in your mysqldump commands, which is good practice anyway. --net_buffer_length of <=100000 have been tested without issue.

serfix is fast but not wildly so. It should run on an average Wordpress database in <1s. In testing, it took ~38s to fix a 250MB .sql file that contained ~940k regexp match/fixes (dumped with --net_buffer_length=75000) on a 2 x 2.4GHz Quad-Core Xeon Mac Pro with 16GB RAM. YMMV.

• Expand test suite
• Remove the second regexp search that is called on every match to find the submatches (this seems to be necessary with the Go standard regexp package, but may not be in the future)
• Better yet, rewrite all regexp operations using a proper lexer

An example of a PHP serialized string is s:20:"http://mywebsite.com";. The number is the character count (with some wrinkles) of the subsequent string, so if the string is changed, PHP will choke when this entry is parsed back into an object.

The need to modify a value (the char count) based on another value (the string), plus the "wrinkes", makes this very difficult (if not impossible) to do with basic unix tools (awk, sed).