[Ericsson AB]

1 Dialyzer

1.1 Introduction

Dialyzer is a static analysis tool that identifies software discrepancies such as type errors, unreachable code, unnecessary tests, etc in single Erlang modules or entire (sets of) applications.

1.2 Using the Dialyzer from the GUI

1.2.1 Choosing the applications or modules

In the "File" window you will find a listing of the current directory. Click your way to the directories/modules you want to add or type the correct path in the entry.

Mark the directories/modules you want to analyze for discrepancies and click "Add". You can either add the .beam and .erl-files directly, or you can add directories that contain these kinds of files. Note that you are only allowed to add the type of files that can be analyzed in the current mode of operation (see below), and that you cannot mix .beam and .erl-files.

1.2.2 The analysis modes

Dialyzer has several modes of analysis. These are controlled by the buttons in the top-middle part of the main window, under "Analysis Options".

The parameters are:

Analyze:
Byte code:
The analysis starts from .beam bytecode files. Whenever the .beam file has been generated with the +debug_info compiler option on, analysis automatically starts from abstract code (via Core Erlang) and the results are identical to those obtained starting from source code.
Source code:
The analysis starts from .erl files.
Granularity:
You can choose to Analyze each module locally, or to make the analysis global over all modules. The default (and recommended mode) is a global analysis, but the module-local variant can be handy when the interfaces between modules are not yet fully implemented and you are simply interested in some quick-and-dirty feedback.
Iteration:
Here you can specify if the analysis should perform a fixpoint iteration over the chosen granularity, or make just one pass.
One Pass:
One pass iteration analyzes all modules in an unspecified order. Each module is still analyzed to fixpoint, but no fixpoint iteration is performed over module boundaries.
Fixpoint:
All specified modules are analyzed until fixpoint. To speed up the iteration, a static call graph for inter-modular calls is constructed after the first pass of the analysis. This call graph is then used to order the modules for the second (and final) run of the analysis.

1.2.3 Controlling the discrepancies reported by the Dialyzer

Under the "Warnings" pull-down menu, there are buttons that control which discrepancies are reported to the user in the "Warnings" window. By clicking on these buttons, one can enable/disable a whole class of warnings. Information about the classes of warnings can be found on the "Warnings" item under the "Help" menu (at the rightmost top corner).

If modules are compiled with inlining, spurious warnings may be emitted. In the "Options" menu you can choose to ignore inline-compiled modules when analyzing byte code. When starting from source code this is not a problem since the inlining is explicitly turned off by Dialyzer. The option causes Dialyzer to suppress all warnings from inline-compiled modules, since there is currently no way for Dialyzer to find what parts of the code have been produced by inlining.

1.2.4 Running the analysis

Once you have chosen the modules or directories you want to analyze, click the "Run" button to start the analysis. If for some reason you want to stop the analysis while it is running, push the "Stop" button.

The information from the analysis will be displayed in the Log and the Warnings windows.

1.2.5 Include directories and macro definitions

When analyzing from source you might have to supply Dialyzer with a list of include directories and macro definitions (as you can do with the erlc flags -I and -D). This can be done either by starting Dialyzer with these flags from the command line as in:

        ./dialyzer -I my_includes -DDEBUG -Dvsn=42 -I one_more_dir
      

or by adding these explicitly using the "Manage Macro Definitions" or "Manage Include Directories" sub-menus in the "Options" menu.

1.2.6 Saving the information on the Log and Warnings windows

In the "File" menu there are options to save the contents of the Log and the Warnings window. Just choose the options and enter the file to save the contents in.

There are also buttons to clear the contents of each window.

1.2.7 Inspecting the inferred types of the analyzed functions

Dialyzer stores the information of the analyzed functions in a Persistent Lookup Table (PLT). After an analysis you can inspect this information. In the PLT menu you can choose to either search the PLT or inspect the contents of the whole PLT. The information is presented in edoc format.

Note

Currently, the information which is displayed is NOT the type signatures of the functions. The return values are the least upper bound of the returned type from the function and the argument types are the least upper bound of the types that the function is called with. In other words, the argument types is not what the function can accept, but rather a description of how the function is used.

We are working on finding the type signatures of the function, and this will (hopefully) be included in a future version of Dialyzer.

1.3 Using the Dialyzer from the command line

See dialyzer(3).

1.4 Using the Dialyzer from Erlang

See dialyzer(3).

1.5 More on the Persistent Lookup Table (PLT)

During setup, a Persistent Lookup Table will automatically be created for the Erlang/OTP libraries specified in dialyzer/src/Makefile. This table will be the starting point for later analyses. At each startup of Dialyzer the validity of the PLT will be checked, and if something has changed in the included libraries a new PLT will be constructed.

To build a PLT of your own favorite files use the --output_plt option to specify where the file containing the PLT should be stored. At later analyses this PLT can be used as the starting point by using the --plt option. Note that the new PLT file also includes the information from the PLT that was used as the starting point

Warning

You should not analyze files that are already included in the PLT. This can lead to unexpected results if some file has changed since the PLT was built. Such dependencies are currently only checked when the PLT was built using the default libraries in dialyzer/src/Makefile.

In an Erlang/OTP system which has been installed, the user typically does not have write permission to the file of the PLT. If Dialyzer later finds that the PLT is not up-to-date, the analysis is aborted with a warning. Until the installed PLT has been updated a temporary PLT can be created and used in the manner described above.

1.6 Feedback and bug reports

At this point, we very much welcome user feedback (even wish-lists!). If you notice something weird, especially if the Dialyzer reports any discrepancy that is a false positive, please send an error report describing the symptoms and how to reproduce them to:

        tobias.lindahl@it.uu.se, kostis@it.uu.se
    

dialyzer 1.7.1
Copyright © 1991-2007 Ericsson AB