JOUST: Journal Of User-Scripted Tests¹

Joust is a testsuite infrastructure consisting of a few components. These either interact directly, or via user scripts. Typically you'll create a file of variable assignments, point at it with the Joust environment variable, then invoke aloy giving it the name of a tester to invoke and a generator program. Aloy reads from the generator's stdout, with each word being a test program to run.

I accidentally wrote this when I realized the test harness I was adding to another project, was a project in its own right. I didn't want to use dejagnu. Grepping for test harnesses found lots of confusing ones. Finally llvm's test harness is not a separate component, and so hard to use elsewhere. You'll notice that Kratos and Ezio have some comonalities with that harness.

Aloy: Apply List, Observe Yield

A testsuite runner that takes a list of test files and command to run. The command is run for each test file. Results are provided in summary and log files. Aloy can run sub jobs in parallel, either communicating with a GNUmake jobserver, or using a specified job limit. Usually the command to run is kratos.

Kratos: Kapture Run And Test Output Safely

A test executor. Looks for RUN lines in the specified text file and executes them, piping stdout &| stderr to checkers. Ability to skip tests.

Ezio: Expect Zero Irregularities Observed

A pattern checker. Looks for CHECK lines in the specified text file and then matches them against the specified file or stdin.

Drake: Dynamic Response And Keyboard Emulation

(Yet to be written)

Joust Tester

A header file and library, to allow testing from your own test program.

CMake Example

Here are CMake snippets for using Joust. Firstly a top-level CMakeLists.txt fragment to determine Joust's location:

# We use Joust for testing.  Where is it?  Rely on the user to tell us.
# This could be done much better.  Perhaps an optional extern?
if (NOT JOUST)
  find_program (ALOY aloy)
  string (REGEX REPLACE "/bin/aloy.*$" "" JOUST ${ALOY})
  if (ALOY EQUAL JOUST)
    set (JOUST)
  endif ()
endif ()

if (JOUST)
  set (JOUST_BINARY_DIR ${JOUST}/bin)
  set (JOUST_INCLUDE_DIR ${JOUST}/include)
  set (JOUST_LIBRARY_DIR ${JOUST}/lib)
  add_custom_target (check)
  message (NOTICE "Joust testsuite:ON")
else ()
  message (WARNING "Joust not available, no testing, use -DJOUST={dir}")
  add_custom_target (check
    echo "Joust not available, configure with -DJOUST={dir}")
endif ()

Then, in a tests subdirectory, some more CMakeLists.txt fragments. Firstly a fragment to run the tests -- I like to make this a different target, dependent on the main 'check' target. Notice the USES_TERMINAL tag, as we write progress to the terminal.

add_custom_target (check-iblis
  srcdir=${CMAKE_CURRENT_SOURCE_DIR} PATH=${JOUST_BINARY_DIR}:$ENV{PATH}
  JOUST=iblis.defs aloy
  -t kratos -o iblis -g ${ZSH} -g ${CMAKE_CURRENT_SOURCE_DIR}/jouster
  COMMENT "Jousting"
  USES_TERMINAL)
add_dependencies (check check-iblis)

You'll need to generate a defs file. Here's such a generator:

file (GENERATE OUTPUT iblis.defs CONTENT
  "testdir=${CMAKE_CURRENT_SOURCE_DIR}
OBJCOPY=${OBJCOPY}
NM=${NM}
GREP=${GREP}
timelimit=60
memlimit=0
cpulimit=60
filelimit=1
SHELL=${ZSH}")

Add to that defs any other variables you need to refer to in the tests.

Finally, create a 'jouster' script in the source directory:

pushd ${0%/*}
setopt nullglob
for subdir in **(/) ; do
    echo $subdir/*.cc
done
popd

And that's it!

Make Example

I used to use Make to build, again with a separate build directory tree. Here are how you can use Joust with Make. A project may consist of multiple source directories, but tests are somewhere under tests subdirectories within this heirarchy.

Here is a Makefile snippet to invoke the testsuite:

ALOY := @ALOY@
# The list of test programs to compile, before invoking tests
TESTS := $(patsubst $(srcdir)/%.cc,%,\
	$(wildcard $(srcdir)/tests/*/*.cc))
TESTDIRS = $(shell cd $(srcdir)/${<D} ; echo *(/))
testdir := $(and $(filter-out /%,$(srcdir)),../)$(srcdir)/tests

# The testsuite.  `+` tells Make to provide a jobserver
check:: tests/myproj.defs $(TESTS)
	+cd ${<D} && srcbuilddir=$(srcdir)/tests JOUST=${<F} \
	  $(ALOY) -t kratos -o myproj -g $(testdir)/jouster $(TESTDIRS)
ifeq ($(firstword $(aloy)),:)
	@echo WARNING: tests were not run as Joust test harness was not found
endif

# The variable definition file.  Add anything you need to refer to from
# kratos or ezio lines.
tests/myproj.defs: tests/Makesub
	echo '# Automatically generated by Make' >$@
	echo "testdir=${testdir}/tests" >>$@
	echo "timelimit=60" >>$@
	echo "memlimit=1" >>$@
	echo "cpulimit=60" >>$@
	echo "filelimit=1" >>$@
	echo "SHELL=$(SHELL)" >>$@
	echo "OBJCOPY=$(OBJCOPY)" >>$@

# Build rule for test programs.
$(TESTS): %: %.o libmyproj.a
	$(CXX) $(LDFLAGS) $< -lmyproj $(LIBS) -o $@

You might notice, that's providing $(testdir)/jouster as a generator program. It happens to be a script in the source directory:

pushd ${0%/*}
setopt nullglob
for subdir in $@ ; do
    echo $subdir/*(.^*)
done
popd

The first test is tests/00-basic/axiom.cc, which is compiled by the above Makefile before the testsuite is invoked. The source file contains the following fragment:

// Check backtrace dtrt
// RUN-SIGNAL:6 $subdir$stem | ezio -p FATAL $src
// FATAL: I am unwell
// FATAL-NEXT: 00-0x{:[0-9a-f]+}
// FATAL: InvokeHCF
// FATAL: main
// FATAL-NEXT: Version
// FATAL-END:

// Check that the ABI is optimal for passing Ref<T>'s the same as T &.
// RUN: $OBJCOPY -Obinary -jidentity1 $subdir$stem.o $tmp.1
// RUN: $OBJCOPY -Obinary -jidentity2 $subdir$stem.o $tmp.2
// RUN: cmp -s $tmp.1 $tmp.2

// RUN-END:

This contains 4 test invocations, one for each RUN: line. The RUN-SIGNAL: line is checking that a failed assert produces a backtrace and exits with a signal. You'll see it's invoking Ezio to check the backtrace, and tells it to use the FATAL: lines for that checking. The next 2 RUN: lines invoke objcopy to extract some information from the object file, writing to two temporary files. The final RUN: line compares those temporaries and we only care it exits with zero.

Aloy: Apply List, Observe Yield

Aloy is the major driver of the testsuite. It is expected that you'll provide some scripts for its use. Typical invocation is:

aloy [options] [tester args --] [generator args]

-C DIR: Change to DIR before doing anything else.
-j COUNT: Fixed job limit
-o STEM Output file stem, defaults to - (stdout/stderr)
-g GEN: Generator program and arguments
-t TESTER Tester program, defaults to kratos

Additional arguments can be passed to the tester program, by using a -- separator after them. The remaining arguments are passed to the generator program. If there is no generator program, they are used directly as the names of tests to run. Comments are introduced with # and extend to end of line.

The number of concurrent jobs to run can be specified with -j, or implicitly using a GNUmake jobserver specified via the MAKEFLAGS environment variable. You may need to prefix the Makerule with +, so that Make knows the rule invokes a jobserver-aware program. Otherwise, although MAKEFLAGS is set, the jobserver is unuseable. Aloy informs you of this happening. Specifying -j overrides any MAKEFILE variable.

Kratos: Kapture Run And Test Output Safely

Kratos scans a source file for marked lines. These are then executed, with the output going to user-specified checker programs. Several separate tests can be specified in a single file, they are executed sequentially.

kratos [options] test-file

-C DIR: Change to DIR before doing anything else.
-D VAR=VALUE: Define variable
-d FILE: Specify file of variable definitions
-o STEM Output file stem, defaults to - (stdout/stderr)
-p PREFIX: Command line prefix, defaults RUN, repeatable

The environment variable $JOUST can be set to specify another file of variable definitions.

RUN: A test pipeline to execute
RUN-SIGNAL: A test pipeline, terminating via a signal
RUN-REQUIRE: A predicate to evaluate
RUN-END: Stop scanning test file

Both RUN and RUN-SIGNAL are similar, except the latter expects the program to terminate via a signal. In both cases the command may specify an exit code or signal, together with ! to indicate not. It doesn't matter what characters appear before RUN on the line (provided there is a non-alphanumeric just before). For instance:

# RUN: true
// RUN:!0 false

The first exits with a zero exit code, and the second exits with a non-zero code.

REQUIRE allows you to disable a following RUN or SIGNAL test. A sequence of REQUIREs all have to succeed, otherwise the test is skipped. REQUIREs only affect one test. You can invert the sense of a REQUIRE with ! in the same way a RUN can be inverted.

The actual command to run is subject to $ expansion, which is similar, but not the same, as shell expansion.

Shell quoting and substitution is different. Arguments are space-separated, use {...} braces to inhibit that — the braces are dropped.
Unlike shell, but like Make, variable expansion is recursive.
Like shell, variable expansion uses $var or ${var}. Spaces in the variable expansion begin new arguments, unless the expansion itself is within braces.
Variables are those specified via a definition file or on the command line.

The program's stdin can be sourced from a file or HERE document and its stdout can be written to one. Input must be on a separate line.

;; 1 RUN: <$testdir/$test ;; 2 RUN: $testdir/$test
;; 3 RUN: $testdir/$test >$tmp-1 ;; 4 RUN: <<line 1 \ ;; 5 RUN: <<line 2 ;; 6 RUN: $testdir/$test

Lines 1 & 2 reads from $testdir/$test, line 3 writes to $tmp-1, lines 4 & 5 are a here document passed to line 6. Notice that the trailing
is literal and does not continue the here line. Here documents are written via a pipe.

Here

Commands may be continued to the next RUN: line by ending with a single \. This cannot appear in the middle of a word though. Regardless of the kind of command started, all continuations must use RUN:.

By default the program's stdout and stderr are buffered and forwarded to kratos's stdout and stderr. So they can be self-checking if wanted. Or the outputs can be piped to checking programs. Often ezio is used to check the output is as expected. Use | to pipe stdout and |& to pipe stderr. You may add these in either order. There can be one or two checkers. If you only have one checker, the other stream is checked to have no output. For example:

// 1 RUN: prog $testdir/$test // 2 RUN: |& ezio $test
// 3 RUN: prog $testdir/$test | ezio -p OUT $test |& ezio $test
// 4 RUN: prog $testdir/$test >$tmp-1 |& ezio $test
// 5 RUN: <$tmp-1 // 6 RUN: ezio -p OUT $test

Here lines 1 & 2 form a single pipeline, with prog's stderr being fed to ezio. Line 3 is similar, but also pipe's prog's stdout to the first invocation of ezio. Lines 4, 5 & 6 do the same checking, but via a temporary file $tmp-1. That file could also be used by a subsequent program under test. ($tmp is an automatically defined variable.)

System resources can be constrained by use of variables:

$cpulimit: Maximum cpu time, in seconds (1 minute).
$memlimit: Maxiumum memory use, in GB (1 GB).
$filelimit: Maximum filesize, in GB (1 GB).
$timelimit: Maximum wall clock time, in seconds (1 minute).

Ezio: Expect Zero Irregularities Observed

Ezio is a pattern matcher. It scans a source file, extracting patterns from it, and then matches the patterns against a test file.

ezio [options] pattern-files+

-C DIR: Change to DIR before doing anything else.
-D VAR=VALUE: Define variable
-d FILE: Specify file of variable definitions
-i INPUT Input file, defaults to - (stdin)
-o STEM Output file stem, defaults to - (stdout/stderr)
-p PREFIX: Command line prefix, defaults CHECK, repeatable

There are several kinds of check patterns. Some are positive matches (& fail if no match is found), others are negative matches (& fail if they match). Positive matches are never checked again, once they match. Negative matches might be checked multiple times (and produce several fails).

CHECK: A synonym for MATCH.
CHECK-MATCH: Match a pattern. Is repeatedly checked until it matches, or we move to a next labelled block.
CHECK-NEXT: Match the next line only. Fails if it doesn't match. If this is the first match, it applies to the first line of the file.
CHECK-DAG: A set of lines may represent a DAG, which implies a partial ordering. This allows the lines of the DAG to be checked.
CHECK-LABEL: Blocks of the input file can be separated with this pattern. If the current pattern does not match, later LABELS are checked to see if we should advance.
CHECK-NOT: A negative match. It is expected not to match, until the next positive match advances.
CHECK-NONE: A negative match that applies to an entire block. Lines that do not make a positive match are expected to not match any of these lines either.
CHECK-NEVER: A negative match that applies to the entire file. As with NONE, lines that fail any positive match are expected to not match these either. Only the first negative match that matches is checked — a NONE and a NEVER that check the same pattern do not trigger twice.
CHECK-OPTION: Set options for the following patterns. Options are:
- matchSol: The pattern is anchored to the start of line.
- matchEol: The pattern is anchored to the end of the line.
- matchLine: The pattern is anchored at both start & end.
- matchSpace: Whitespace must match exactly.
- xfail: The pattern is xfailed — expected to fail.
Options are space or , separated. They may be preceded by ! to turn the option off. matchLine is equivalent to specifying both matchSol and matchEol. An xfail only applies to the next pattern, unlike the other options it resets.

Patterns are literal matches, with the following extensions:

Leading and trailing whitespace is elided. If you must match such space, use a regexp escpe.
A leading ^ anchors a particlar pattern at the start of line (regardless of matchSol).
A trailing $ anchors at the end of line. (The options are useful if you need to anchor a whole set of lines.)
Whitespace matches any sequence of whitespace.
Variable expansions are denoted by $VAR or ${VAR}. These are matched literally.
To match a literal $ use ${}.
Captures are denoted by {VAR:REGEXP}, which if the match is successful set the variable to the text the regexp matched. Posix extended regexps are used.
Plain regexps are simply captures without a variable: {:REGEXP}.
The regexp can contain a variable expansion, which is matched literally (not asa regexp itself). The above $VAR or ${VAR} syntax does that. As before ${} obtains a literal $, but you might need to escape that with \, because of its meaning in a REGEXP.
Patterns can contain variables whose value has not yet been determined. These patterns do not match at that point. If Ezio can determine the variable can never have a value set until too late, an error is given. This is particularly signficant with DAG matching, where later patterns of the DAG contain expansion captured from earlier patterns.

DRAKE: Dynamic Response And Keyboard Emulation

Libjoust

You can use libjoust directly, to write unit tests.

// Kratos runline
// RUN: $subdir$stem

#include "joust.hh"  // Yeah, I like .hh, bite me.

int main ()
{
  Joust::Tester tester;

  int error_code = PerformSomeTest ();
  if (error_code)
    // Emit arbitrary explanatory text to the log stream
    tester.Log () << "Foo test produced " << error_code << '\n';

  // Inform of results (source location will be included)
  tester.Result (Joust::Tester::PassFail (!error_code)) << "Foo test";
}

If you're driving this from Kratos, the test program should return zero, except in the case of configuration error or other failure external to the test itself.

Commonalities

As you might expect for a set of related programs, they share some semantics and syntax.

Variables

Variables can be specfied in:

-Dvar=val command line options
-d file file specified on the command line
File specified in a $JOUST environment variable.

They are initialized in that order, and the first initializer wins. These are unrelated to environment variables. (Although environment variables can influence tests, because they are visible to the programs being tested.) Typically setup code creates a file of values and initialize $JOUST to point at it.

All the programs expect a $testdir variable to be specified pointing at the test directory. They automatically prepend it to certain arguments. The following variables are automatically created:

$test: The test file being tested.
$stem: The basename of $test, stripping both directory and suffix components.
$subdir: The directory fragment of $test including a final /. If there is no directory component, an empty string is assigned.
$tmp: A temporary name constructed from $test by replacing every / with - and appending .tmp.

Input and Output

Ezio, by default, reads the file to check from stdin. You can use -i FILE to provide a file to read. If FILE is -, it signifies stdin anyway.

Aloy, Kratos & Ezio report their results to stdout and stderr. stdout provides summary results — just pass/fail of each tests. stderr provides logging, which along with the pas/fail information shows informative test output so you can see why it failed, and information about how to reexecute just that test, either inside the test harness, or from the shell. All of them allow you to redirect that output to two files with the same basename with the -o STEM option. Summary resuls go to STEM.sum and logging output goes to STEM.log. Usually you'll give Aloy a -o mytests option.

Note that if output is going to stdout and stderr, and they refer to the same file, output is generally undecipherably duplicated. Use the shell to redirect to separate files (>sum 2>log).

Future

Kratos offloading to a remote execution system. Add $wrapper variable or something?
Kratos copying files to/from remote system. Add $cpto $cp from variables along with RUN-AUX: or similar.
Kratos iteration over a set of flags. Add RUN-ITERATE: along with ability to defer some toplevel variable expansion to runtime. RUN-REQUIRE inside a loop would continue to the next iteration of the loop.

Building Joust

Building Joust is reasonably straight forwards. You need a C++20 compiler, as I use some features of C++20. Use the CMake idiom of building in a subdirectory called 'build{something}'

Here's a recipe:

git clone git@github.com:urnathan/joust.git
cd joust
mkdir build
cd build
cmake ..
make
make check

1: or 'Journal Of Utterly Stupid Tests'

Name		Name	Last commit message	Last commit date
Latest commit History 184 Commits
extern		extern
gaige		gaige
include		include
joust		joust
progs		progs
tests		tests
.clang-format		.clang-format
.gitignore		.gitignore
.gitmodules		.gitmodules
CMakeLists.txt		CMakeLists.txt
LICENSE		LICENSE
README.md		README.md
cfg.h.in		cfg.h.in
gdbinit.in		gdbinit.in

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

JOUST: Journal Of User-Scripted Tests¹

CMake Example

Make Example

Aloy: Apply List, Observe Yield

Kratos: Kapture Run And Test Output Safely

Ezio: Expect Zero Irregularities Observed

DRAKE: Dynamic Response And Keyboard Emulation

Libjoust

Commonalities

Variables

Input and Output

Future

Building Joust

About

Releases

Packages

Languages

License

urnathan/joust

Folders and files

Latest commit

History

Repository files navigation

JOUST: Journal Of User-Scripted Tests1

CMake Example

Make Example

Aloy: Apply List, Observe Yield

Kratos: Kapture Run And Test Output Safely

Ezio: Expect Zero Irregularities Observed

DRAKE: Dynamic Response And Keyboard Emulation

Libjoust

Commonalities

Variables

Input and Output

Future

Building Joust

About

Resources

License

Stars

Watchers

Forks

Releases

Packages 0

Languages

JOUST: Journal Of User-Scripted Tests¹

Packages