[lilypond.git] / Documentation / contributor / regressions.itexi

@c -*- coding: utf-8; mode: texinfo; -*-
@node Regression tests
@chapter Regression tests

@menu
* Introduction to regression tests::
* Precompiled regression tests::
* Compiling regression tests::
* Regtest comparison::
* Finding the cause of a regression::
* Memory and coverage tests::
* MusicXML tests::
* Grand Regression Test Checking::
@end menu


@node Introduction to regression tests
@section Introduction to regression tests

LilyPond has a complete suite of regression tests that are used
to ensure that changes to the code do not break existing behavior.
These regression tests comprise small LilyPond snippets that test
the functionality of each part of LilyPond.

Regression tests are added when new functionality is added to
LilyPond.
We do not yet have a policy on when it is appropriate to add or
modify a regtest when bugs are fixed.  Individual developers
should use their best judgement until this is clarified during the
@ref{Grand Organization Project (GOP)}.

The regression tests are compiled using special @code{make}
targets.  There are three primary uses for the regression
tests.  First, successful completion of the regression tests means
that LilyPond has been properly built.  Second, the output of the
regression tests can be manually checked to ensure that
the graphical output matches the description of the intended
output.  Third, the regression test output from two different
versions of LilyPond can be automatically compared to identify
any differences.  These differences should then be manually
checked to ensure that the differences are intended.

Regression tests (@qq{regtests}) are available in precompiled form
as part of the documentation.  Regtests can also be compiled
on any machine that has a properly configured LilyPond build
system.


@node Precompiled regression tests
@section Precompiled regression tests

@subheading Regression test output

As part of the release process, the regression tests are run
for every LilyPond release.  Full regression test output is
available for every stable version and the most recent development
version.

Regression test output is available in HTML and PDF format.  Links
to the regression test output are available at the developer's
resources page for the version of interest.

The latest stable version of the regtests is found at:

@example
@uref{http://lilypond.org/doc/stable/input/regression/collated-files.html}
@end example

The latest development version of the regtests is found at:

@example
@uref{http://lilypond.org/doc/latest/input/regression/collated-files.html}
@end example


@subheading Regression test comparison

Each time a new version is released, the regtests are
compiled and the output is automatically compared with the
output of the previous release.  The result of these
comparisons is archived online:

@example
@uref{http://lilypond.org/test/}
@end example

Checking these pages is a very important task for the LilyPond project.
You are invited to report anything that looks broken, or any case
where the output quality is not on par with the previous release,
as described in @rweb{Bug reports}.

@warning{ The special regression test
@file{test-output-distance.ly} will always show up as a
regression.  This test changes each time it is run, and serves to
verify that the regression tests have, in fact, run.}


@subheading What to look for

The test comparison shows all of the changes that occurred between
the current release and the prior release.  Each test that has a
significant difference in output is displayed, with the old
version on the left and the new version on the right.

Regression tests whose output is the same for both versions are
not shown in the test comparison.

@itemize
@item
Images: green blurs in the new version show the approximate
location of elements in the old version.

There are often minor adjustments in spacing which do not indicate
any problem.

@item
Log files: show the difference in command-line output.

The main thing to examine are any changes in page counts -- if a
file used to fit on 1 page but now requires 4 or 5 pages,
something is suspicious!

@item
Profile files: give information about
TODO?  I don't know what they're for.

@end itemize

@warning{
The automatic comparison of the regtests checks the LilyPond
bounding boxes.  This means that Ghostscript changes and changes
in lyrics or text are not found.
}

@node Compiling regression tests
@section Compiling regression tests

Developers may wish to see the output of the complete regression
test suite for the current version of the source repository
between releases.  Current source code is available; see
@ref{Working with source code}.

For regression testing @code{../configure} should be run with the
@code{--disable-optimising} option.  Then you will need
to build the LilyPond binary; see @ref{Compiling LilyPond}.

Uninstalling the previous LilyPond version is not necessary, nor is
running @code{make install}, since the tests will automatically be
compiled with the LilyPond binary you have just built in your source
directory.

From this point, the regtests are compiled with:

@example
make test
@end example

If you have a multi-core machine you may want to use the @option{-j}
option and @var{CPU_COUNT} variable, as
described in @ref{Saving time with CPU_COUNT}.
For a quad-core processor the complete command would be:

@example
make -j5 CPU_COUNT=5 test
@end example

The regtest output will then be available in
@file{input/regression/out-test}.
@file{input/regression/out-test/collated-examples.html}
contains a listing of all the regression tests that were run,
but none of the images are included.  Individual images are
also available in this directory.

The primary use of @samp{make@tie{}test} is to verify that the
regression tests all run without error.  The regression test
page that is part of the documentation is created only when the
documentation is built, as described in @ref{Generating documentation}.
Note that building the documentation requires more installed components
than building the source code, as described in
@ref{Requirements for building documentation}.


@node Regtest comparison
@section Regtest comparison

Before modified code is committed to @code{master} (via @code{staging}),
a regression test
comparison must be completed to ensure that the changes have
not caused problems with previously working code.  The comparison
is made automatically upon compiling the regression test suite
twice.

@enumerate

@item
Run @code{make} with current git master without any of your changes.

@item
Before making changes to the code, establish a baseline for the comparison by
going to the @file{lilypond-git/build/} directory and running:

@example
make test-baseline
@end example

@item
Make your changes, or apply the patch(es) to consider.

@item
Compile the source with @samp{make} as usual.

@item
Check for unintentional changes to the regtests:

@example
make check
@end example

After this has finished, a regression test comparison will be
available (relative to the current @file{build/} directory) at:

@example
out/test-results/index.html
@end example

For each regression test that differs between the baseline and the
changed code, a regression test entry will be displayed.  Ideally,
the only changes would be the changes that you were working on.
If regressions are introduced, they must be fixed before
committing the code.

@warning{
The special regression test @file{test-output-distance.ly} will always
show up as a regression.  This test changes each time it is run, and
serves to verify that the regression tests have, in fact, run.}

@item
If you are happy with the results, then stop now.

If you want to continue programming, then make any additional code
changes, and continue.

@item
Compile the source with @samp{make} as usual.

@item
To re-check files that differed between the initial
@samp{make@tie{}test-baseline} and your post-changes
@samp{make@tie{}check}, run:

@example
make test-redo
@end example

This updates the regression list at @file{out/test-results/index.html}.
It does @emph{not} redo @file{test-output-distance.ly}.

@item
When all regressions have been resolved, the output list will be empty.

@item
Once all regressions have been resolved, a final check should be completed
by running:

@example
make test-clean
make check
@end example

This cleans the results of the previous @samp{make@tie{}check}, then does the
automatic regression comparison again.  

@end enumerate

@advanced{
Once a test baseline has been established, there is no need to run it again
unless git master changed. In other words, if you work with several branches
and want to do regtests comparison for all of them, you can
@code{make test-baseline} with git master, checkout some branch,
@code{make} and @code{make check} it, then switch to another branch,
@code{make test-clean}, @code{make} and @code{make check} it without doing
@code{make test-baseline} again.}


@node Finding the cause of a regression
@section Finding the cause of a regression

Git has special functionality to help tracking down the exact
commit which causes a problem.  See the git manual page for
@code{git bisect}.  This is a job that non-programmers can do,
although it requires familiarity with git, ability to compile
LilyPond, and generally a fair amount of technical knowledge.  A
brief summary is given below, but you may need to consult other
documentation for in-depth explanations.

Even if you are not familiar with git or are not able to compile
LilyPond you can still help to narrow down the cause of a
regression simply by downloading the binary releases of different
LilyPond versions and testing them for the regression.  Knowing
which version of LilyPond first exhibited the regression is
helpful to a developer as it shortens the @code{git bisect}
procedure.

Once a problematic commit is identified, the programmers' job is
much easier.  In fact, for most regression bugs, the majority of
the time is spent simply finding the problematic commit.

More information is in @ref{Regression tests}.

@subheading git bisect setup

We need to set up the bisect for each problem we want to
investigate.

Suppose we have an input file which compiled in version 2.13.32,
but fails in version 2.13.38 and above.

@enumerate
@item
Begin the process:

@example
git bisect start
@end example

@item
Give it the earliest known bad tag:

@example
git bisect bad release/2.13.38-1
@end example

(you can see tags with: @code{git tag} )

@item
Give it the latest known good tag:

@example
git bisect good release/2.13.32-1
@end example

You should now see something like:
@example
Bisecting: 195 revisions left to test after this (roughly 8 steps)
[b17e2f3d7a5853a30f7d5a3cdc6b5079e77a3d2a] Web: Announcement
update for the new @qq{LilyPond Report}.
@end example

@end enumerate

@subheading git bisect actual

@enumerate

@item
Compile the source:

@example
make
@end example

@item
Test your input file:

@example
out/bin/lilypond test.ly
@end example

@item
Test results?

@itemize
@item
Does it crash, or is the output bad?  If so:

@example
git bisect bad
@end example

@item
Does your input file produce good output?  If so:

@example
git bisect good
@end example

@end itemize

@item
Once the exact problem commit has been identified, git will inform
you with a message like:

@example
6d28aebbaaab1be9961a00bf15a1ef93acb91e30 is the first bad commit
%%% ... blah blah blah ...
@end example

If there is still a range of commits, then git will automatically
select a new version for you to test.  Go to step #1.

@end enumerate

@subheading Recommendation: use two terminal windows

@itemize
@item
One window is open to the @code{build/} directory, and alternates
between these commands:

@example
make
out/bin/lilypond test.ly
@end example

@item
One window is open to the top source directory, and alternates
between these commands:

@example
git bisect good
git bisect bad
@end example

@end itemize


@node Memory and coverage tests
@section Memory and coverage tests

In addition to the graphical output of the regression tests, it is
possible to test memory usage and to determine how much of the source
code has been exercised by the tests.

@subheading Memory usage

For tracking memory usage as part of this test, you will need
GUILE CVS; especially the following patch:
@smallexample
@uref{http://www.lilypond.org/vc/old/gub.darcs/patches/guile-1.9-gcstats.patch}.
@end smallexample

@subheading Code coverage

For checking the coverage of the test suite, do the following

@example
./scripts/auxiliar/build-coverage.sh
@emph{# uncovered files, least covered first}
./scripts/auxiliar/coverage.py  --summary out-cov/*.cc
@emph{# consecutive uncovered lines, longest first}
./scripts/auxiliar/coverage.py  --uncovered out-cov/*.cc
@end example


@node MusicXML tests
@section MusicXML tests


LilyPond comes with a complete set of regtests for the
@uref{http://www.musicxml.org/,MusicXML} language.  Originally
developed to test @samp{musicxml2ly}, these regression tests
can be used to test any MusicXML implementation.

The MusicXML regression tests are found at
@file{input/regression/musicxml/}.

The output resulting from running these tests
through @samp{musicxml2ly} followed by @samp{lilypond} is
available in the LilyPond documentation:

@example
@uref{http://lilypond.org/doc/latest/input/regression/musicxml/collated-files}
@end example


@node Grand Regression Test Checking
@section Grand Regression Test Checking

@subheading What is this all about?

Regression tests (usually abbreviated "regtests") is a collection
of @file{.ly} files used to check whether LilyPond is working correctly.
Example: before version 2.15.12 breve noteheads had incorrect width,
which resulted in collisions with other objects.  After the issue was fixed,
a small @file{.ly} file demonstrating the problem was added to the regression
tests as a proof that the fix works.  If someone will accidentally break
breve width again, we will notice this in the output of that regression test.

@subheading How can I help?

We ask you to help us by checking one or two regtests from time to time.
You don't need programming skills to do this, not even LilyPond skills -
just basic music notation knowledge; checking one regtest takes less than
a minute.  Simply go here:

@example
@uref{http://www.philholmes.net/lilypond/regtests/}
@end example

@subheading Some tips on checking regtests

@subsubheading Description text

The description should be clear even for a music beginner.
If there are any special terms used in the description,
they all should be explained in our @rglosnamed{Top, Music Glossary}
or @rinternalsnamed{Top, Internals Reference}.
Vague descriptions (like "behaves well", "looks reasonable") shouldn't be used.

@ignore
this may be useful for advanced regtest checking
@subsubheading Is regtest straightforward and systematic?

Unfortunately some regtests are written poorly.  A good regtest should be
straightforward: it should be obvious what it checks and how.  Also, it
usually shouldn't check everything at once.  For example it's a bad idea to test
accidental placement by constucting one huge chord with many suspended notes
and loads of accidentals.  It's better to divide such problem into a series
of clearly separated cases.
@end ignore