This document describes the changes made to the Dialyzer application.
Add a --fullpath option to Dialyzer
This change adds a --fullpath option to Dialyzer, which makes the warning messages contain the full path of the corresponding file.
Original patch submitted by Magnus Henoch (legoscia) on 15/9/2010 and cooked to death in the 'pu' branch all this time.
The patch was essentially correct and most of it has been used as is, but there have been some changes to make the code slightly prettier, avoid some code duplication, and add documentation to dialyzer's doc files and to its help message.
Own Id: OTP-9098
Fix warnings about guards containing not
The wording of warnings about unsatisfiable guards that used 'not' was incorrect (the 'not' was not mentioned and it appeared as "Guard test is_atom(atom()) can never succeed") (thanks to Stavros Aronis).
Own Id: OTP-9099
Version 2.4.2 (in Erlang/OTP R14B02) ------------------------------------ - Added --fullpath option to display files with warnings with their full file names (thanks to Magnus Henoch for the original patch). - Better handling of 'and'/'or'/'not' guards that generate warnings (thanks to Stavros Aronis). - Better blame assignment for cases when a function's spec is erroneous (thanks to Stavros Aronis). - More descriptive warnings when a tuple/record pattern contains subterms that violate the declared types of record fields (thanks to Matthias Lang for the test case and for Stavros Aronis for the actual fix).
Own Id: OTP-9126
Add spec to dialyzer_cl_parse:get_lib_dir/1
Own Id: OTP-9129
Test suites for Dialyzer
This is a transcription of most of the cvs.srv.it.uu.se:/hipe repository dialyzer_tests into test suites that use the test server framework.
See README for information on how to use the included scripts for modifications and updates.
When testing Dialyzer it's important that several OTP modules are included in the plt. The suites takes care of that too.
Own Id: OTP-9116
- Fixed pretty rare infinite loop when refining the types of an SCC whose functions all returned none() (thanks to Stavros Aronis).
- Fixed pretty rare crash when taking the infimum of two tuple_sets.
Own Id: OTP-8979
- Added ability to supply multiple PLTs for the analysis (option --plts). Currently these PLTs must be independent (i.e., no module appears in more than one PLT) and there must not include files with module name clashes.
- Strengthened and streamlined hard-coded type information for some BIFs and key library functions.
Own Id: OTP-8962
Eliminated warnings for auto-imported BIF clashes.
Own Id: OTP-8840
Various changes to dialyzer-related files for R14.
- Dialyzer properly supports the new attribute -export_type and checks that remote types only refer to exported types. A warning is produced if some files/applications refer to types defined in modules which are neither in the PLT nor in the analyzed applications.
- Support for detecting data races involving whereis/1 and unregister/1.
- More precise identification of the reason(s) why a record construction violates the types declared for its fields.
- Fixed bug in the handling of the 'or' guard.
- Better handling of the erlang:element/2 BIF.
- Complete handling of Erlang BIFs.
Own Id: OTP-8699
Much better support for opaque types (thanks to Manouk Manoukian).
Added support for recursive types (experimental).
Added support for parameterized modules.
Dialyzer now warns when -specs state that a function returns some type when in fact it does not.
Added
Fixed minor bug in the dialyzer script allowing the --wx option to bring up the wx-based GUI regardless of its placement in the options list.
Options --apps and -Wrace_conditions, which were added in the previous version, are now properly documented in the manual.
Own Id: OTP-8464
The documentation is now built with open source tools (xsltproc and fop) that exists on most platforms. One visible change is that the frames are removed.
Own Id: OTP-8201
Dialyzer can statically detect some kinds of data races in Erlang programs. Use the new option -Wrace_conditions to enable the race analysis. The technique is described in a paper which is available at: http://www.it.uu.se/research/group/hipe/dialyzer/publications/races.pdf
Added support for packages (thanks to Maria Christakis).
There has been a major change in the default mode of Dialyzer. Previously, the default mode was the GUI, while now it is the command line. As a result of this change, it is now possible to analyze a set of files/dirs with the command:
In other words, the -c (--command-line) option is no longer necessary, though it will be retained for some time for backwards compatibility. To start dialyzer's GUI use either of the following commands:
There is a new option --apps which allows the user to easily refer to Erlang/OTP applications and include them in the analysis or in the building of the PLT. For example, we recommend building the PLT with:
The new option can also take absolute file names as well as applications. Note that the application versions that will be included in the PLT are those that correspond to the Erlang/OTP system which is used.
Dialyzer has a new wxWidgets based GUI (thanks to Elli Frangaki) for platforms where the wx application is available.
Own Id: OTP-8300
There is a major addition to the capabilities of dialyzer, worthy of bumping the version number. Starting with this version, dialyzer not only accepts but also properly processes remote types (i.e., types of the form ModuleName:TypeName()). Previous dialyzer versions only accepted this notation in -type and -spec declarations, but effectively ignored its information by mapping remote types to the type any(). In contrast, starting with this version, remote types are used in the analysis and are also stored in the PLTs. (This addition changes the format of PLTs and requires rebuilding any PLTs created by an older dialyzer version.) Note that dialyzer will complain and abort the analysis of a set of modules if it needs to process a remote type without a definition (either because the module does not include a definition of the type or the module is not included in the analysis). We may relax this restriction in a future version.
Fixed minor issue with dialyzer:run/1 crashing (upon its return) when used for adding type information to an existing PLT.
Fixed minor but quite annoying issues in dialyzer's GUI.
Own Id: OTP-8187
Fixed problem with type inference going into an infinite loop when analyzing a strongly connected component of functions that do not return but also contain an erroneous call which makes them fail and be assigned the type none() instead of the type unit().
More accurate type information for some BIFs and library files.
Introduced boolean() as the `official' name for the type was so far known as bool(). The latter is still accepted as well as boolean().
Own Id: OTP-8037
Has better handling of opaque types.
The handling of UTF segments of bitstreams has been significantly strengthened and revised. In all probability, now it is correct.
Own Id: OTP-7958
The analysis accepts opaque type declarations and detects violations of opaqueness of terms of such types. Starting with R13, many Erlang/OTP standard libraries (array, dict, digraph, ets, gb_sets, gb_trees, queue, and sets) contain opaque type declarations of their main data types. Dialyzer will spit out warnings in code that explicitly depends on the structure of these terms.
Added support for handling UTF segments in bitstreams and for detecting obvious type errors in these segments. Warning: This code is not terribly tested though since there are very few Erlang programs which use Unicode-based binaries - not surprising since this is a new language feature of R13.
Strengthened the discrepancy identification when testing for equality and matching between terms of different types. This detects more bugs in code.
Added warning for M:F(...) calls where M is not a module term and F is not an atom. Previously, such calls where found to fail but the reason for the failure was not reported.
Added a convenient shorthand for the --no_check_plt option (-n).
Added the --dump_callgraph option for dumping the callgraph of all files that are analyzed into a specified file. The callgraph either be dumped in raw format, in .dot format, or converted to a .ps (postscript) file. Note that in large callgraphs the generated postscript file might not be interpretable by Ghostview. (Thanks to Ilya Khlopotov for the initial version of this functionality.)
Own Id: OTP-7864
Added the
Changed the phrasing of some warnings so that they do not cause confusion to some users and correspond better to reality.
Own Id: OTP-7632
Minor updates.
Own Id: OTP-7522
There is new
Own Id: OTP-7386
The return type of the Erlang interface dialyzer:run/1 has changed to only return a list of warnings. If something goes wrong dialyzer dies with an exception.
The handling of the PLT is now more flexible. There is no longer any default PLT included with OTP. Please consult the manual for the changes.
Own Id: OTP-7389
Dialyzer's analysis is from now on exclusively based
on success typings. In particular, support for options
Better and more aggressive handling of type information in records.
Dialyzer has a new warning option
Some warning options (
Own Id: OTP-7241
The warnings returned by the Erlang interface now contains a tag describing the type of warning.
*** POTENTIAL INCOMPATIBILITY ***
Own Id: OTP-7134
Use of success typings is now default, is robust and detects significantly more errors than in previous versions.
Dialyzer now accepts and takes into account type information in record declarations and in contracts -- see related paper in Erlang'07 workshop. Various OTP applications (e.g. stdlib and kernel) are partially annotated with appropriate contracts specifying types information for their functions.
The type previously known as unit() has been renamed to no_return(). Its use in a contract as the return type of a function now silences off the "function has no local return" dialyzer warning without use of the corresponding option.
Own Id: OTP-6997
Minor Makefile changes.
Own Id: OTP-6689
Dialyzer can now use success typings to find discrepancies. As a consequence significantly more discrepancies are detected. The downside is that the analysis takes about 2 to 2.5 times longer than in previous versions, and may also result in some warnings that might be harder to interpret even though they are correct. This has been a major change, worth of increasing the version number.
New command-line options:
--succ_typings Use the success typings analysis.
--dataflow (default) Use the previous analysis.
The new type unit() has been introduced to handle the return type of non-terminating functions such as servers.
Dialyzer's code server uses a compressed representation and the analysis requires significantly less memory than in previous versions.
Own Id: OTP-6736
Dialyzer (in the modes where either source or BEAM code which contains debug_info is analyzed) spits out line information for all discrepancies. This has been a major change, worth of increasing the version number.
We warn users that *considerably* more discrepancies are identified by this version of Dialyzer compared with previous ones and applications. If, for some reason, warnings generated by previous versions of Dialyzer are preferable, the command line option --old_style can be employed.
Own Id: OTP-6546
Dialyzer handles term comparison operators more precisely and is able to identify more discrepancies in 'if' or 'case' statements with comparisons.
Dialyzer has more precise type information for many standard OTP functions.
Own Id: OTP-6547
Updated the chapter "More on the Persistent Lookup Table
(PLT)" in Dialyzer User's Guide and added information on
how to use Dialyzer from Erlang to
Own Id: OTP-6361
New options
Improvements in the analysis (meaning that this version can find more discrepancies than the previous version).
Own Id: OTP-6421
Dialyzer's building of PLT is based on a different type inference algorithm. More specifically, Dialyzer uses inference of refined success typings to infer function prototypes. As a result, Dialyzer bases its analysis on a significantly more powerful basis and thus is able to detect more discrepancies. The downside is that building the PLT is a considerably slower process. We will work on improving that.
Dialyzer takes into account the BEAM compiler
directive
Dialyzer's default initial PLT now also includes "mnesia".
Own Id: OTP-6304
Improvements in PLT management.
Own Id: OTP-6128
Some minor changes.
Some minor changes.
Changes for Dialyzer to work with Erlang/OTP R10B-10.
Dialyzer's analysis is significantly faster as it uses a global function call-graph for obtaining type signatures for all analyzed functions.
Small changes for Dialyzer to work with Erlang/OTP R10B-5.
Fixed a small buglet in the analysis; this affected one of HiPE's files.
Modified setup script for execution under Cygwin (patch by Vlad Dumitrescu).
Added command line option --no_warn_on_inline.
Dialyzer now explicitly warns when modules with the same name but from different dirs are being analyzed (contribution by Ulf Wiger).
Fixed a number of false positives that Dialyzer 1.2.0 was spitting out.
Requires the presence of an Erlang/OTP R10B-4 system.
Dialyzer is significantly (approx 40%) faster since it now uses 'ets' rather than 'dets' for its PLT.
Slightly improved the precision of the analysis.
In the GUI version, Dialyzer now reports the list of modules that should be included in the modules to analyze in order to possibly improve the accuracy of the reported results.
Some more information is displayed when calling a function or closure with arguments of the wrong type.
The record guard now allows discrepancies involving tuples that are known to be records to be displayed as #rec{} rather than {'rec',_,...,_}
Added -q option which makes the command-line version of Dialyzer a bit more silent.
Dialyzer works with the open source and commercial versions of Erlang/OTP R10B-2 on all platforms (i.e., HiPE support is not a prerequisite anymore).
Whenever a .beam file contains abstract code (i.e., has been compiled with the +debug_info option), the analysis starts from this code rather than from BEAM bytecode -- this makes the results identical to those obtained by analysis starting from source code. (This is a contribution from Bjorn Gustavsson - Thanks!)
Added -pa command line option for easier handling of -include_lib() directives in source code.
Includes all changes added in v1.1.1; see below. The "Add Recursively" functionality is also available in the command-line mode (-r option).
Fixed problems using the shell script on Solaris machines.
Fixed small inconsistencies in Dialyzer's documentation and help.
The command-line mode of Dialyzer spits discrepancies in stdout by default and returns a numeric value indicating its exit status.
Added "Add Recursively" button contributed by Martin Bjorklund (thanks!).
Dialyzer works with the open source version of Erlang/OTP R10B-1 (on platforms where HiPE support is available) and it does not require installation of a customized Erlang/OTP system anymore.
Dialyzer comes with an extensive command-line interface, in addition to an enhanced GUI.
Analysis can start from source code (through Core Erlang) as well as from BEAM bytecode.
Dialyzer finds considerably more discrepancies in Erlang code than previous versions.
Added ability to selectively turn on/off different warning categories.
Fixed major bug of v1.0.0 which caused the analysis to stop prematurely (i.e., without properly reaching fixpoint) when analyzing more than one module. This did not affect correctness of the results, but lost many opportunities to detect discrepancies.
Performance improvements in the analysis.
Initial Dialyzer version: Analysis starts from BEAM bytecode only and only a GUI version is available. Dialyzer reports discrepancies in the use of the "sequential" part of Erlang. To be used, Dialyzer requires the presence of a specific snapshot of Erlang/OTP which can be obtained from Dialyzer's homepage.