This is an essay by Jim Blandy on maintaining ChangeLog entries. Although Subversion generates its ChangeLogs from svn log data, instead of keeping independent ChangeLog files, most of the advice below is as applicable to cvs log messages as to ChangeLog entries. Maintaining the ChangeLog ========================= A project's ChangeLog provides a history of development. Comments in the code should explain the code's present state, but ChangeLog entries should explain how and when it got that way. The ChangeLog must show: * the relative order in which changes entered the code, so you can see the context in which a change was made, and * the date at which the change entered the code, so you can relate the change to outside events, like branch cuts, code freezes, and releases. In the case of CVS, these refer to when the change was committed, because that is the context in which other developers will see the change. Every change to the sources should have a ChangeLog entry. The value of the ChangeLog becomes much less if developers cannot rely on its completeness. Even if you've only changed comments, write an entry that says, "Doc fix." The only changes you needn't log are small changes that have no effect on the source, like formatting tweaks. In order to keep the ChangeLog a manageable size, at the beginning of each year, the ChangeLog should be renamed to "ChangeLog-YYYY", and a fresh ChangeLog file started. How to write ChangeLog entries ------------------------------ ChangeLog entries should be full sentences, not sentence fragments. Fragments are more often ambiguous, and it takes only a few more seconds to write out what you mean. Fragments like `New file' or `New function' are acceptable, because they are standard idioms, and all further details should appear in the source code. The log entry should mention every file changed. It should also mention by name every function, variable, macro, makefile target, grammar rule, etc. you changed. However, there are common-sense exceptions: * If you have made a change which requires trivial changes throughout the rest of the program (e.g., renaming a variable), you needn't name all the functions affected. * If you have rewritten a file completely, the reader understands that everything in it has changed, so your log entry may simply give the file name, and say "Rewritten". In general, there is a tension between making entries easy to find by searching for identifiers, and wasting time or producing unreadable entries by being exhaustive. Use your best judgement --- and be considerate of your fellow developers. Group ChangeLog entries into "paragraphs", separated by blank lines. Each paragraph should be a set of changes that accomplish a single goal. Independent changes should be in separate paragraphs. For example: 1999-03-24 Stan Shebs * configure.host (mips-dec-mach3*): Use mipsm3, not mach3. Attempt to sort out SCO-related configs. * configure.host (i[3456]86-*-sysv4.2*): Use this instead of i[3456]86-*-sysv4.2MP and i[3456]86-*-sysv4.2uw2*. (i[3456]86-*-sysv5*): Recognize this. * configure.tgt (i[3456]86-*-sco3.2v5*, i[3456]86-*-sco3.2v4*): Recognize these. Even though this entry describes two changes to `configure.host', they're in separate paragraphs, because they're unrelated changes. The second change to `configure.host' is grouped with another change to `configure.tgt', because they both serve the same purpose. Also note that the author has kindly recorded his overall motivation for the paragraph, so we don't have to glean it from the individual changes. The header line for the ChangeLog entry should have the format shown above. If you are using an old version of Emacs (before 20.1) that generates entries with more verbose dates, consider using `etc/add-log.el', from the GDB source tree. If you are using vi, consider using the macro in `etc/add-log.vi'. Both of these generate entries in the newer, terser format. One should never need the ChangeLog to understand the current code. If you find yourself writing a significant explanation in the ChangeLog, you should consider carefully whether your text doesn't actually belong in a comment, alongside the code it explains. Here's an example of doing it right: 1999-02-23 Tom Tromey * cplus-dem.c (consume_count): If `count' is unreasonable, return 0 and don't advance input pointer. And then, in `consume_count' in `cplus-dem.c': while (isdigit ((unsigned char)**type)) { count *= 10; count += **type - '0'; /* A sanity check. Otherwise a symbol like `_Utf390_1__1_9223372036854775807__9223372036854775' can cause this function to return a negative value. In this case we just consume until the end of the string. */ if (count > strlen (*type)) { *type = save; return 0; } This is why a new function, for example, needs only a log entry saying "New Function" --- all the details should be in the source. Avoid the temptation to abbreviate filenames or function names, as in this example (mostly real, but slightly exaggerated): * gdbarch.[ch] (gdbarch_tdep, gdbarch_bfd_arch_info, gdbarch_byte_order, {set,}gdbarch_long_bit, {set,}gdbarch_long_long_bit, {set,}gdbarch_ptr_bit): Corresponding functions. This makes it difficult for others to search the ChangeLog for changes to the file or function they are interested in. For example, if you searched for `set_gdbarch_long_bit', you would not find the above entry, because the writer used CSH-style globbing to abbreviate the list of functions. If you gave up, and made a second pass looking for gdbarch.c, you wouldn't find that either. Consider your poor readers, and write out the names. ChangeLogs and the CVS log -------------------------- CVS maintains its own logs, which you can access using the `cvs log' command. This duplicates the information present in the ChangeLog, but binds each entry to a specific revision, which can be helpful at times. However, the CVS log is no substitute for the ChangeLog files. * CVS provides no easy way to see the changes made to a set of files in chronological order. They're sorted first by filename, not by date. * Unless you put full ChangeLog paragraphs in your CVS log entries, it's difficult to pull together changes that cross several files. * CVS doesn't segregate log entries for branches from those for the trunk in any useful way. In some circumstances, though, the CVS log is more useful than the ChangeLog, so we maintain both. When you commit a change, you should provide appropriate text in both the ChangeLog and the CVS log. It is not necessary to provide CVS log entries for ChangeLog changes, since it would simply duplicate the contents of the file itself. Writing ChangeLog entries for merges ------------------------------------ Revision management software like CVS can introduce some confusion when writing ChangeLog entries. For example, one might write a change on a branch, and then merge it into the trunk months later. In that case, what position and date should the developer use for the ChangeLog entry --- that of the original change, or the date of the merge? The principles described at the top need to hold for both the original change and the merged change. That is: * On the branch (or trunk) where the change is first committed, the ChangeLog entry should be written as normal, inserted at the top of the ChangeLog and reflecting the date the change was committed to the branch (or trunk). * When the change is then merged (to the trunk, or to another branch), the ChangeLog entry should have the following form: 1999-03-26 Jim Blandy Merged change from foobar_20010401_branch: 1999-03-16 Keith Seitz [...] In this case, "Jim Blandy" is doing the merge on March 26; "Keith Seitz" is the original author of the change, who committed it to `foobar_20010401_branch' on March 16. As shown here, the entry for the merge should be like any other change --- inserted at the top of the ChangeLog, and stamped with the date the merge was committed. It should indicate the origin of the change, and provide the full text of the original entry, indented to avoid being confused with a true log entry. Remember that people looking for the merge will search for the original changelog text, so it's important to preserve it unchanged. For the merge entry, we use the merge date, and not the original date, because this is when the change appears on the trunk or branch this ChangeLog documents. Its impact on these sources is independent of when or where it originated. This approach preserves the structure of the ChangeLog (entries appear in order, and dates reflect when they appeared), but also provides full information about changes' origins.