5.13. Debugging the compiler


5.13.1. Dumping out compiler intermediate structures


Causes the output from each of flags starting with “-ddump”, to be dumped to a file or files. If you want to have all the output from one single flag saved to one file, use -ddump-file-prefix=⟨str⟩ (see descriptions below). Otherwise, the output will go to several files, including one for non-module specific and several for module specific. The suffix of a dump file depends on the flag turned on, for instance, output from -ddump-simpl will end up in prefix.dump-simpl.


Set the prefix of the filenames used for debugging output. For example, -ddump-file-prefix=Foo will cause the output from -ddump-simpl to be dumped to Foo.dump-simpl.


When compiling Main.hs with profiling and without this will now produce Main.p.dump-simpl and Main.dump-simpl instead of overwriting the output of one way with the output of another.


Dump error messages as JSON documents. This is intended to be consumed by external tooling. A good way to use it is in conjunction with -ddump-to-file.


Print out each pass name, its runtime and heap allocations as it happens. Note that this may come at a slight performance cost as the compiler will be a bit more eager in forcing pass results to more accurately account for their costs.

Two types of messages are produced: Those beginning with *** do denote the beginning of a compilation phase whereas those starting with !!! mark the end of a pass and are accompanied by allocation and runtime statistics.


For each module, show some simple statistics about which info tables have IPE information, and how many info tables with IPE information each closure type has. This is useful, for example, for verifying that STACK info tables are being appropriately omitted or included from the info table map.


Show statistics on the usage of fast strings by the compiler.


Dump the whole FastString table when finished. Consider using -ddump-file-prefix=⟨str⟩ to dump it into a file.


Debugging output is in one of several “styles.” Take the printing of types, for example. In the “user” style (the default), the compiler’s internal ideas about types are presented in Haskell source-level syntax, insofar as possible. In the “debug” style (which is the default for debugging output), the types are printed in with explicit foralls, and variables have their unique-id attached (so you can check for things that look the same but aren’t). This flag makes debugging output appear in the more verbose debug style.


Show allocation and runtime statistics for various stages of compilation. Allocations are measured in bytes. Timings are measured in milliseconds.

GHC is a large program consisting of a number of stages. You can tell GHC to dump information from various stages of compilation using the -ddump-⟨pass⟩ flags listed below. Note that some of these tend to produce a lot of output. You can prevent them from clogging up your standard output by passing -ddump-to-file. Front-end

These flags dump various information from GHC’s frontend. This includes the parser and interface file reader.


Dump parser output


Dump parser output as a syntax tree


Include comments in the parser. Useful in combination with -ddump-parsed-ast.


Make the interface loader be real chatty about what it is up to. Type-checking and renaming

These flags dump various information from GHC’s typechecker and renamer.


Make the type checker be real chatty about what it is up to.


Make the renamer be real chatty about what it is up to.


Make the pattern match exhaustiveness checker be real chatty about what it is up to.


Make the constraint solver be real chatty about what it is up to.


Print out summary of what kind of information the renamer had to bring in.


Dump renamer output


Dump renamer output as a syntax tree


Dump typechecker output. Note that this hides a great deal of detail by default; you might consider using this with -fprint-typechecker-elaboration.


Dump typechecker output as a syntax tree


Dump the hie file syntax tree if we are generating extended interface files


Dump Template Haskell expressions that we splice in, and what Haskell code the expression evaluates to.


Dump expansions of all top-level Template Haskell splices into module.th.hs for each file module.hs.


Dump a type signature for each value defined at the top level of the module. The list is sorted alphabetically. Using -dppr-debug dumps a type signature for all the imported and system-defined things as well; useful for debugging the compiler.


Dump derived instances Core representation and simplification

These flags dump various phases of GHC’s Core-to-Core pipeline. This begins with the desugarer and includes the simplifier, worker-wrapper transformation, the rule engine, the specialiser, the strictness/occurrence analyser, and a common subexpression elimination pass.


Dump output of the call arity analysis pass (-fcall-arity).


Print a one-line summary of the size of the Core program at the end of the optimisation pipeline.


Dump desugarer output. -ddump-ds dumps the output after the very simple optimiser has run (which discards a lot of clutter and hence is a sensible default. -ddump-ds-preopt shows the output after desugaring but before the very simple optimiser.


Dump output of the exitification pass (-fexitification), which tries to pull out code out of recursive functions.


Show the output of each iteration of the simplifier (each run of the simplifier has a maximum number of iterations, normally 4).


Dump statistics about how many of each kind of transformation took place. If you add -dppr-debug you get more detailed information.


Dump trace messages from various functions of the simplifier. Produces quite a lot of output.


Show the output of the intermediate Core-to-Core pass. (lots of output!) So: when we’re really desperate:

% ghc -noC -O -ddump-simpl -dverbose-core2core -dcore-lint Foo.hs

Dump output of specialisation pass


Dumps all rewrite rules specified in this module; see Controlling what’s going on in rewrite rules.


Dumps the names of all rules that fired in this module


Dumps detailed information about all rules that fired in this module


This flag is useful for debugging why a rule you expect to be firing isn’t.

Rules are filtered by the user provided string, a rule is kept if a prefix of its name matches the string. The pass then checks whether any of these rules could apply to the program but which didn’t fire for some reason. For example, specifying -drule-check=SPEC will check whether there are any applications which might be subject to a rule created by specialisation.


This flag is useful for debugging why a definition is not inlined.

When a string is passed to this flag we report information about all functions whose name shares a prefix with the string.

For example, if you are inspecting the core of your program and you observe that foo is not being inlined. You can pass -dinline-check foo and you will see a report about why foo is not inlined.


Dump simplifier output (Core-to-Core passes)


Dumps inlinings performed by the simplifier.


Dumps all inlinings considered by the simplifier, even those ultimately not performed. This output includes various information that the simplifier uses to determine whether the inlining is beneficial.


Dump demand analysis output.

See -fstrictness for the syntax and semantics of demand annotations.


Dump top-level demand signatures as produced by demand analysis.

See -fstrictness for the syntax and semantics of demand annotations.


Dump Constructed Product Result analysis output


Dump Constructed Product Result signatures


Dump common subexpression elimination (CSE) pass output


Dump worker/wrapper split output


Dump “occurrence analysis” output


Dump output of Core preparation pass


Dump output of LateCC pass after cost centres have been added.


Print the view patterns that are commoned. STG representation

These flags dump various phases of GHC’s STG pipeline.


Show the output of CoreToStg pass.


Show the output of the intermediate STG-to-STG pass. (lots of output!)


Show the output of the unarise pass.


Show the output of the STG after Stg2Stg. This is the result after applying the Stg2Stg optimization passes.


Show the output of the tag inference pass.


Show the output of the last STG pass before we generate Cmm.


Alias for -ddump-stg-from-core. Deprecated in favor of more explicit flags: -ddump-stg-from-core, -ddump-stg-final, etc. C-- representation

These flags dump various phases of GHC’s C-- pipeline.


Dump output from main C-- pipeline stages. In case of .cmm compilation this also dumps the result of file parsing. Not included are passes run by the chosen backend. Currently only the NCG backends runs additional passes ( -ddump-opt-cmm ).

Cmm dumps don’t include unreachable blocks since we print blocks in reverse post-order.


If used in conjunction with -ddump-to-file, writes dump output from main C-- pipeline stages to files (each stage per file).


Dump the result of STG-to-C-- conversion


Dump the “raw” C--.


Dump the results of the C-- control flow optimisation pass.


Dump the results of the C-- pass responsible for adding instrumentation added by -fcmm-thread-sanitizer.


Dump the results of the C-- Common Block Elimination (CBE) pass.


Dump the results of the C-- switch lowering pass.


Dump the results of the C-- proc-point analysis pass.


Dump the results of the C-- stack layout pass.


Dump the results of the C-- sinking pass.


Dump the results of the C-- CAF analysis pass.


Dump the results of the C-- proc-point map pass.


Dump the results of the C-- proc-point splitting pass.


Dump the results of the C-- info table augmentation pass.


Dump the results of the CPS pass.


Dump the result of the C-- pipeline processing


Dumps the CFG with weights used by the new block layout code. Each CFG is dumped in dot format graph making it easy to visualize them. LLVM code generator


LLVM code from the LLVM code generator C code generator

Shortdesc:Dump C code produced by the C (unregisterised) backend. Native code generator

These flags dump various stages of the native code generator’s pipeline, which starts with C-- and produces native assembler.


Dump the results of C-- to C-- optimising passes performed by the NCG.


Alias for -ddump-cmm-opt


Dump (virtual) register conflicts (“interferences”) from the graph coloring register allocator (-fregs-graph).


Dump the initial assembler output produced from C--.


Dump the result of the register liveness pass.


Dump the result of the register allocation pass.


Dump the build/spill stages of the -fregs-graph register allocator.


Dump statistics from the register allocator.


Dump the final assembly produced by the native code generator.


Dump the final JavaScript code produced by the JavaScript code generator. Miscellaneous backend dumps

These flags dump various bits of information from other backends.


Dump byte-code objects (BCOs) produced for the GHC’s byte-code interpreter.


Dump generated debug information (DWARF) produced with the -g flag.


Trace runtime type inference done by various interpreter commands.


Dump foreign export stubs.


Dump the code instrumented by HPC (Observing Code Coverage).


An alias for -ddump-ticked.


Dump a mapping of modules to where they come from, and how:

  • (hidden module): Module is hidden, and thus will never be available for import.
  • (unusable module): Module is unavailable because the package is unusable.
  • (hidden package): This module is in someone’s exported-modules list, but that package is hidden.
  • (exposed package): Module is available for import.
  • (reexport by <PACKAGES>): This module is available from a reexport of some set of exposed packages.
  • (hidden reexport by <PACKAGES>): This module is available from a reexport of some set of hidden packages.
  • (package flag): This module export comes from a package flag.

5.13.2. Formatting dumps


In error messages, expressions are printed to a certain “depth”, with subexpressions beyond the depth replaced by ellipses. This flag sets the depth. Its default value is 5.


Set the width of debugging output. Use this if your code is wrapping too much. For example: -dppr-cols=200.


Print single alternative case expressions as though they were strict let expressions. This is helpful when your code does a lot of unboxing.


Print values of type Word# and Word64# (but not values of type Int# and Int64#) in hexadecimal instead of decimal. The hexadecimal is zero-padded to make the length of the representation a power of two. For example: 0x0A0A##, 0x000FFFFF##, 0xC##. This flag may be helpful when you are producing a bit pattern that to expect to work correctly on a 32-bit or a 64-bit architecture. Dumping hexadecimal literals after optimizations and constant folding makes it easier to confirm that the generated bit pattern is correct.


Suppress any unsolicited debugging output. When GHC has been built with the DEBUG option it occasionally emits debug output of interest to developers. The extra output can confuse the testing framework and cause bogus test failures, so this flag is provided to turn it off.

5.13.3. Suppressing unwanted information

Core dumps contain a large amount of information. Depending on what you are doing, not all of it will be useful. Use these flags to suppress the parts that you are not interested in.


Suppress everything that can be suppressed, except for unique ids as this often makes the printout ambiguous. If you just want to see the overall structure of the code, then start here.


Suppress “ticks” in the pretty-printer output.


Suppress the printing of uniques. This may make the printout ambiguous (e.g. unclear where an occurrence of ‘x’ is bound), but it makes the output of two compiler runs have many fewer gratuitous differences, so you can realistically apply diff. Once diff has shown you where to look, you can try again without -dsuppress-uniques


Suppress extended information about identifiers where they are bound. This includes strictness information and inliner templates. Using this flag can cut the size of the core dump in half, due to the lack of inliner templates


Suppress the printing of the stable unfolding of a variable at its binding site.


Suppress the printing of module qualification prefixes. This is the Data.List in Data.List.length.


Suppress the printing of timestamps. This makes it easier to diff dumps.


Suppress the printing of type signatures.


Suppress the printing of type applications.


Suppress the printing of type coercions.


Suppress the printing of variable kinds


Suppress the printing of closure free variable lists in STG output


Suppress the printing of core size stats per binding


default: enabled

Disabling this will annoate certain stg arguments with their prim rep.

5.13.4. Checking for consistency

Implies:-dcore-lint, -dstg-lint, -dcmm-lint, -dasm-lint, -fllvm-fill-undef-with-garbage, -debug

Turn on various heavy-weight intra-pass sanity-checking measures within GHC and its runtime system. Notably, this does not include -falignment-sanitisation as it incurs a rather hefty runtime cost.


Turn on heavyweight intra-pass sanity-checking within GHC, at Core level. (It checks GHC’s sanity, not yours.)


Turn on linearity checking in GHC. Currently, some optimizations in GHC might not preserve linearity and there are valid programs that fail Linear Core Lint. In the near future, this option will be removed and folded into normal Core Lint.


Ditto for STG level.


Ditto for C-- level.


Turn on intra-pass sanity-checking within GHC, at the code generator level.


Instructs the LLVM code generator to fill dead STG registers with garbage instead of undef in calls. This makes it easier to catch subtle code generator and runtime system bugs (e.g. see #11487).


Compile with alignment checks for all info table dereferences. This can be useful when finding pointer tagging issues.


Align functions to multiples of the given value. Only valid values are powers of two.

-fproc-alignment=64 can be used to limit alignment impact on performance as each function will start at a cache line. However forcing larger alignments in general reduces performance.


GHC generates case expressions without a default alternative in some cases:

  • When the demand analysis thinks that the scrutinee does not return (i.e. a bottoming expression)
  • When the scrutinee is a GADT and its type rules out some constructors, and others constructors are already handled by the case expression.

With this flag GHC generates a default alternative with error in these cases. This is helpful when debugging demand analysis or type checker bugs which can sometimes manifest as segmentation faults.


Typically primops operations like writeArray# exhibit unsafe behavior, relying on the user to perform any bounds checking. This flag instructs the code generator to instrument such operations with bound checking logic which aborts the program when an out-of-bounds access is detected.

Note that this is only intended to be used as a debugging measure, not as the primary means of catching out-of-bounds accesses.


This enables generation of ThreadSanitizer <https://github.com/google/sanitizers/wiki/ThreadSanitizerCppManual> instrumentation of memory accesses. Requires use of -fsanitize=thread or similar when compiling and linking.

5.13.5. Checking for determinism


Start UniqSupply allocation from ⟨s⟩.


Set the increment for the generated Unique’s to ⟨i⟩.

This is useful in combination with -dinitial-unique=⟨s⟩ to test if the generated files depend on the order of Unique’s.

Some interesting values:

  • -dinitial-unique=0 -dunique-increment=1 - current sequential UniqSupply
  • -dinitial-unique=16777215 -dunique-increment=-1 - UniqSupply that generates in decreasing order
  • -dinitial-unique=1 -dunique-increment=PRIME - where PRIME big enough to overflow often - nonsequential order

5.13.6. Other


This avoids generating Typeable-related bindings for modules and types. This is useful when debugging because it gives smaller modules and dumps, but the compiler will panic if you try to use Typeable instances of things that you built with this flag.


When tag inference tells as a specific value is supposed to be tagged then generate code to check this at runtime. If the check fails the program will be terminated. This helps narrowing down if an issue is due to tag inference if things go wrong. Which would otherwise be quite difficult.