View Source cover - The Coverage Analysis Tool
Introduction
The module cover
provides a set of functions for coverage analysis of Erlang
programs, counting how many times each executable line
is executed.
Coverage analysis can be used to verify test cases, making sure all relevant code is covered, and can be helpful when looking for bottlenecks in the code.
Getting Started With Cover
Example
Assume that a test case for the following program should be verified:
-module(channel).
-behaviour(gen_server).
-export([start_link/0,stop/0]).
-export([alloc/0,free/1]). % client interface
-export([init/1,handle_call/3,terminate/2]). % callback functions
start_link() ->
gen_server:start_link({local,channel}, channel, [], []).
stop() ->
gen_server:call(channel, stop).
%%%-Client interface functions-------------------------------------------
alloc() ->
gen_server:call(channel, alloc).
free(Channel) ->
gen_server:call(channel, {free,Channel}).
%%%-gen_server callback functions----------------------------------------
init(_Arg) ->
{ok,channels()}.
handle_call(stop, _Client, Channels) ->
{stop,normal,ok,Channels};
handle_call(alloc, _Client, Channels) ->
{Ch,Channels2} = alloc(Channels),
{reply,{ok,Ch},Channels2};
handle_call({free,Channel}, _Client, Channels) ->
Channels2 = free(Channel, Channels),
{reply,ok,Channels2}.
terminate(_Reason, _Channels) ->
ok.
%%%-Internal functions---------------------------------------------------
channels() ->
[ch1,ch2,ch3].
alloc([Channel|Channels]) ->
{Channel,Channels};
alloc([]) ->
false.
free(Channel, Channels) ->
[Channel|Channels].
The test case is implemented as follows:
-module(test).
-export([s/0]).
s() ->
{ok,Pid} = channel:start_link(),
{ok,Ch1} = channel:alloc(),
ok = channel:free(Ch1),
ok = channel:stop().
Preparation
First of all, Cover must be started. This spawns a process which owns the Cover database where all coverage data will be stored.
1> cover:start().
{ok,<0.90.0>}
To include other nodes in the coverage analysis, use
cover:start/1
. All cover-compiled modules will then be loaded on all
nodes, and data from all nodes will be summed up when analysing. For
simplicity this example only involves the current node.
Before any analysis can take place, the involved modules must be
cover-compiled. This means that some extra information is added to
the module before beging compiled into a binary and
loaded. The source file of the module is
not affected and no .beam
file is created.
2> cover:compile_module(channel).
{ok,channel}
Each time a function in the cover-compiled module channel
is called,
information about the call will be added to the Cover database. Run the test
case:
3> test:s().
ok
Cover analysis is performed by examining the contents of the Cover database. The
output is determined by two parameters, Level
and Analysis
. Analysis
is
either coverage
or calls
and determines the type of the analysis. Level
is
either module
, function
, clause
, or line
and determines the level of the
analysis.
Coverage Analysis
Analysis of type coverage
is used to find out how much of the code has been
executed and how much has not been executed. Coverage is represented by a tuple
{Cov,NotCov}
, where Cov
is the number of executable lines that have been
executed at least once and NotCov
is the number of executable lines that have
not been executed.
If the analysis is made on module level, the result is given for the entire
module as a tuple {Module,{Cov,NotCov}}
:
4> cover:analyse(channel, coverage, module).
{ok,{channel,{14,1}}}
For channel
, the result shows that 14 lines in the module are covered but one
line is not covered.
If the analysis is made on function level, the result is given as a list of
tuples {Function,{Cov,NotCov}}
, one for each function in the module. A
function is specified by its module name, function name and arity:
5> cover:analyse(channel, coverage, function).
{ok,[{{channel,start_link,0},{1,0}},
{{channel,stop,0},{1,0}},
{{channel,alloc,0},{1,0}},
{{channel,free,1},{1,0}},
{{channel,init,1},{1,0}},
{{channel,handle_call,3},{5,0}},
{{channel,terminate,2},{1,0}},
{{channel,channels,0},{1,0}},
{{channel,alloc,1},{1,1}},
{{channel,free,2},{1,0}}]}
For channel
, the result shows that the uncovered line is in the function
channel:alloc/1
.
If the analysis is made on clause level, the result is given as a list of tuples
{Clause,{Cov,NotCov}}
, one for each function clause in the module. A clause is
specified by its module name, function name, arity and position within the
function definition:
6> cover:analyse(channel, coverage, clause).
{ok,[{{channel,start_link,0,1},{1,0}},
{{channel,stop,0,1},{1,0}},
{{channel,alloc,0,1},{1,0}},
{{channel,free,1,1},{1,0}},
{{channel,init,1,1},{1,0}},
{{channel,handle_call,3,1},{1,0}},
{{channel,handle_call,3,2},{2,0}},
{{channel,handle_call,3,3},{2,0}},
{{channel,terminate,2,1},{1,0}},
{{channel,channels,0,1},{1,0}},
{{channel,alloc,1,1},{1,0}},
{{channel,alloc,1,2},{0,1}},
{{channel,free,2,1},{1,0}}]}
For channel
, the result shows that the uncovered line is in the second clause
of channel:alloc/1
.
Finally, if the analysis is made on line level, the result is given as a list of
tuples {Line,{Cov,NotCov}}
, one for each executable line in the source code. A
line is specified by its module name and line number.
7> cover:analyse(channel, coverage, line).
{ok,[{{channel,9},{1,0}},
{{channel,12},{1,0}},
{{channel,17},{1,0}},
{{channel,20},{1,0}},
{{channel,25},{1,0}},
{{channel,28},{1,0}},
{{channel,31},{1,0}},
{{channel,32},{1,0}},
{{channel,35},{1,0}},
{{channel,36},{1,0}},
{{channel,39},{1,0}},
{{channel,44},{1,0}},
{{channel,47},{1,0}},
{{channel,49},{0,1}},
{{channel,52},{1,0}}]}
For channel
, the result shows that the uncovered line is line number 49.
Call Statistics
Analysis of type calls
is used to find out how many times something has been
called and is represented by an integer Calls
.
If the analysis is made on module level, the result is given as a tuple
{Module,Calls}
. Here Calls
is the total number of calls to functions in the
module:
8> cover:analyse(channel, calls, module).
{ok,{channel,12}}
For channel
, the result shows that a total of twelve calls have been made to
functions in the module.
If the analysis is made on function level, the result is given as a list of
tuples {Function,Calls}
. Here Calls
is the number of calls to each function:
9> cover:analyse(channel, calls, function).
{ok,[{{channel,start_link,0},1},
{{channel,stop,0},1},
{{channel,alloc,0},1},
{{channel,free,1},1},
{{channel,init,1},1},
{{channel,handle_call,3},3},
{{channel,terminate,2},1},
{{channel,channels,0},1},
{{channel,alloc,1},1},
{{channel,free,2},1}]}
For channel
, the result shows that handle_call/3
is the most called function
in the module (three calls). All other functions have been called once.
If the analysis is made on clause level, the result is given as a list of tuples
{Clause,Calls}
. Here Calls
is the number of calls to each function clause:
10> cover:analyse(channel, calls, clause).
{ok,[{{channel,start_link,0,1},1},
{{channel,stop,0,1},1},
{{channel,alloc,0,1},1},
{{channel,free,1,1},1},
{{channel,init,1,1},1},
{{channel,handle_call,3,1},1},
{{channel,handle_call,3,2},1},
{{channel,handle_call,3,3},1},
{{channel,terminate,2,1},1},
{{channel,channels,0,1},1},
{{channel,alloc,1,1},1},
{{channel,alloc,1,2},0},
{{channel,free,2,1},1}]}
For channel
, the result shows that all clauses have been called once, except
the second clause of channel:alloc/1
which has not been called at all.
Finally, if the analysis is made on line level, the result is given as a list of
tuples {Line,Calls}
. Here Calls
is the number of times each line has been
executed:
11> cover:analyse(channel, calls, line).
{ok,[{{channel,9},1},
{{channel,12},1},
{{channel,17},1},
{{channel,20},1},
{{channel,25},1},
{{channel,28},1},
{{channel,31},1},
{{channel,32},1},
{{channel,35},1},
{{channel,36},1},
{{channel,39},1},
{{channel,44},1},
{{channel,47},1},
{{channel,49},0},
{{channel,52},1}]}
For channel
, the result shows that all lines have been executed once, except
line number 49 which has not been executed at all.
Analysis to File
A line level calls analysis of channel
can be written to a file using
cover:analyse_to_file/1
:
12> cover:analyse_to_file(channel).
{ok,"channel.COVER.out"}
The function creates a copy of channel.erl
where it for each executable line
is specified how many times that line has been executed. The output file is
called channel.COVER.out
.
File generated from /Users/bjorng/git/otp/channel.erl by COVER 2024-03-20 at 13:25:04
****************************************************************************
| -module(channel).
| -behaviour(gen_server).
|
| -export([start_link/0,stop/0]).
| -export([alloc/0,free/1]). % client interface
| -export([init/1,handle_call/3,terminate/2]). % callback functions
|
| start_link() ->
1..| gen_server:start_link({local,channel}, channel, [], []).
|
| stop() ->
1..| gen_server:call(channel, stop).
|
| %%%-Client interface functions-------------------------------------------
|
| alloc() ->
1..| gen_server:call(channel, alloc).
|
| free(Channel) ->
1..| gen_server:call(channel, {free,Channel}).
|
| %%%-gen_server callback functions----------------------------------------
|
| init(_Arg) ->
1..| {ok,channels()}.
|
| handle_call(stop, _Client, Channels) ->
1..| {stop,normal,ok,Channels};
|
| handle_call(alloc, _Client, Channels) ->
1..| {Ch,Channels2} = alloc(Channels),
1..| {reply,{ok,Ch},Channels2};
|
| handle_call({free,Channel}, _Client, Channels) ->
1..| Channels2 = free(Channel, Channels),
1..| {reply,ok,Channels2}.
|
| terminate(_Reason, _Channels) ->
1..| ok.
|
| %%%-Internal functions---------------------------------------------------
|
| channels() ->
1..| [ch1,ch2,ch3].
|
| alloc([Channel|Channels]) ->
1..| {Channel,Channels};
| alloc([]) ->
0..| false.
|
| free(Channel, Channels) ->
1..| [Channel|Channels].
Conclusion
By looking at the results from the analyses, it can be deduced that
the test case does not cover the case when all channels are allocated
and test.erl
should be extended accordingly. Incidentally, when the
test case is corrected a bug in channel
will be discovered.
When the Cover analysis is ready, Cover is stopped and all cover-compiled
modules are unloaded. The code for channel
is now
loaded as usual from a .beam
file in the current path.
13> code:which(channel).
cover_compiled
14> cover:stop().
ok
15> code:which(channel).
"./channel.beam"
Miscellaneous
Performance
Execution of code in cover-compiled modules is slower and more memory consuming than for regularly compiled modules. As the Cover database contains information about each executable line in each cover-compiled module, performance decreases proportionally to the size and number of the cover-compiled modules.
To improve performance when analysing cover results it is possible to do multiple calls to analyse and analyse_to_file at once. You can also use the async_analyse_to_file convenience function.
Executable Lines
Cover uses the concept of executable lines, which is code lines containing
an executable expression such as a matching or a function call. A blank line or
a line containing a comment, function head or pattern in a case
or receive
statement is not executable.
In the example below, lines number 2, 4, 6, 8, and 11 are executable lines:
1: is_loaded(Module, Compiled) ->
2: case get_file(Module, Compiled) of
3: {ok,File} ->
4: case code:which(Module) of
5: ?TAG ->
6: {loaded,File};
7: _ ->
8: unloaded
9: end;
10: false ->
11: false
12: end.
Code Loading Mechanism
When a module is cover-compiled, it is also loaded using the normal code loading
mechanism of Erlang. This means that if a cover-compiled module is re-loaded
during a Cover session, for example using c(Module)
, it will no longer be
cover-compiled.
Use cover:is_compiled/1
or code:which/1
to see whether or not a
module is cover-compiled (and still loaded).
When Cover is stopped, all cover-compiled modules are unloaded.