1 <Sect1 id="options-debugging">
2 <Title>Debugging the compiler
6 <IndexTerm><Primary>debugging options (for GHC)</Primary></IndexTerm>
10 HACKER TERRITORY. HACKER TERRITORY.
14 <Sect2 id="replacing-phases">
15 <Title>Replacing the program for one or more phases.
19 <IndexTerm><Primary>GHC phases, changing</Primary></IndexTerm>
20 <IndexTerm><Primary>phases, changing GHC</Primary></IndexTerm>
21 You may specify that a different program be used for one of the phases
22 of the compilation system, in place of whatever the driver <Literal>ghc</Literal> has
23 wired into it. For example, you might want to try a different
25 <Literal>-pgm<phase-code><program-name></Literal><IndexTerm><Primary>-pgm<phase><stuff>
26 option</Primary></IndexTerm> option to <Literal>ghc</Literal> will cause it to use <Literal><program-name></Literal>
27 for phase <Literal><phase-code></Literal>, where the codes to indicate the phases are:
33 <ColSpec Align="Left" Colsep="0">
34 <ColSpec Align="Left" Colsep="0">
37 <Entry><Emphasis>code</Emphasis> </Entry>
38 <Entry><Emphasis>phase</Emphasis> </Entry>
44 <Entry> literate pre-processor </Entry>
49 <Entry> C pre-processor (if -cpp only) </Entry>
54 <Entry> Haskell compiler </Entry>
59 <Entry> C compiler</Entry>
64 <Entry> assembler </Entry>
69 <Entry> linker </Entry>
74 <Entry> Makefile dependency generator </Entry>
85 <Sect2 id="forcing-options-through">
86 <Title>Forcing options to a particular phase.
90 <IndexTerm><Primary>forcing GHC-phase options</Primary></IndexTerm>
94 The preceding sections describe driver options that are mostly
95 applicable to one particular phase. You may also <Emphasis>force</Emphasis> a
96 specific option <Literal><option></Literal> to be passed to a particular phase
97 <Literal><phase-code></Literal> by feeding the driver the option
98 <Literal>-opt<phase-code><option></Literal>.<IndexTerm><Primary>-opt<phase><stuff>
99 option</Primary></IndexTerm> The codes to indicate the phases are the same as in the
104 So, for example, to force an <Literal>-Ewurble</Literal> option to the assembler, you
105 would tell the driver <Literal>-opta-Ewurble</Literal> (the dash before the E is
110 Besides getting options to the Haskell compiler with <Literal>-optC<blah></Literal>,
111 you can get options through to its runtime system with
112 <Literal>-optCrts<blah></Literal><IndexTerm><Primary>-optCrts<blah> option</Primary></IndexTerm>.
116 So, for example: when I want to use my normal driver but with my
117 profiled compiler binary, I use this script:
121 exec /local/grasp_tmp3/simonpj/ghc-BUILDS/working-alpha/ghc/driver/ghc \
122 -pgmC/local/grasp_tmp3/simonpj/ghc-BUILDS/working-hsc-prof/hsc \
132 <Sect2 id="dumping-output">
133 <Title>Dumping out compiler intermediate structures
137 <IndexTerm><Primary>dumping GHC intermediates</Primary></IndexTerm>
138 <IndexTerm><Primary>intermediate passes, output</Primary></IndexTerm>
145 <Term><Literal>-noC</Literal>:</Term>
148 <IndexTerm><Primary>-noC option</Primary></IndexTerm>
149 Don't bother generating C output <Emphasis>or</Emphasis> an interface file. Usually
150 used in conjunction with one or more of the <Literal>-ddump-*</Literal> options; for
151 example: <Literal>ghc -noC -ddump-simpl Foo.hs</Literal>
156 <Term><Literal>-hi</Literal>:</Term>
159 <IndexTerm><Primary>-hi option</Primary></IndexTerm>
160 <Emphasis>Do</Emphasis> generate an interface file. This would normally be used in
161 conjunction with <Literal>-noC</Literal>, which turns off interface generation;
162 thus: <Literal>-noC -hi</Literal>.
167 <Term><Literal>-dshow-passes</Literal>:</Term>
170 <IndexTerm><Primary>-dshow-passes option</Primary></IndexTerm>
171 Prints a message to stderr as each pass starts. Gives a warm but
172 undoubtedly misleading feeling that GHC is telling you what's
178 <Term><Literal>-ddump-<pass></Literal>:</Term>
181 <IndexTerm><Primary>-ddump-<pass> options</Primary></IndexTerm>
182 Make a debugging dump after pass <Literal><pass></Literal> (may be common enough to
183 need a short form…). You can get all of these at once (<Emphasis>lots</Emphasis> of
184 output) by using <Literal>-ddump-all</Literal>, or most of them with <Literal>-ddump-most</Literal>.
185 Some of the most useful ones are:
192 <Term><Literal>-ddump-parsed</Literal>:</Term>
200 <Term><Literal>-ddump-rn</Literal>:</Term>
208 <Term><Literal>-ddump-tc</Literal>:</Term>
216 <Term><Literal>-ddump-deriv</Literal>:</Term>
224 <Term><Literal>-ddump-ds</Literal>:</Term>
232 <Term><Literal>-ddump-spec</Literal>:</Term>
235 output of specialisation pass
240 <Term><Literal>-ddump-rules</Literal>:</Term>
243 dumps all rewrite rules (including those generated by the specialisation pass)
248 <Term><Literal>-ddump-simpl</Literal>:</Term>
251 simplifer output (Core-to-Core passes)
256 <Term><Literal>-ddump-usagesp</Literal>:</Term>
259 UsageSP inference pre-inf and output
264 <Term><Literal>-ddump-cpranal</Literal>:</Term>
272 <Term><Literal>-ddump-stranal</Literal>:</Term>
275 strictness analyser output
280 <Term><Literal>-ddump-workwrap</Literal>:</Term>
283 worker/wrapper split output
288 <Term><Literal>-ddump-occur-anal</Literal>:</Term>
291 `occurrence analysis' output
296 <Term><Literal>-ddump-stg</Literal>:</Term>
299 output of STG-to-STG passes
304 <Term><Literal>-ddump-absC</Literal>:</Term>
307 <Emphasis>un</Emphasis>flattened Abstract C
312 <Term><Literal>-ddump-flatC</Literal>:</Term>
315 <Emphasis>flattened</Emphasis> Abstract C
320 <Term><Literal>-ddump-realC</Literal>:</Term>
323 same as what goes to the C compiler
328 <Term><Literal>-ddump-asm</Literal>:</Term>
331 assembly language from the native-code generator
339 <IndexTerm><Primary>-ddump-all option</Primary></IndexTerm>
340 <IndexTerm><Primary>-ddump-most option</Primary></IndexTerm>
341 <IndexTerm><Primary>-ddump-parsed option</Primary></IndexTerm>
342 <IndexTerm><Primary>-ddump-rn option</Primary></IndexTerm>
343 <IndexTerm><Primary>-ddump-tc option</Primary></IndexTerm>
344 <IndexTerm><Primary>-ddump-deriv option</Primary></IndexTerm>
345 <IndexTerm><Primary>-ddump-ds option</Primary></IndexTerm>
346 <IndexTerm><Primary>-ddump-simpl option</Primary></IndexTerm>
347 <IndexTerm><Primary>-ddump-cpranal option</Primary></IndexTerm>
348 <IndexTerm><Primary>-ddump-workwrap option</Primary></IndexTerm>
349 <IndexTerm><Primary>-ddump-rules option</Primary></IndexTerm>
350 <IndexTerm><Primary>-ddump-usagesp option</Primary></IndexTerm>
351 <IndexTerm><Primary>-ddump-stranal option</Primary></IndexTerm>
352 <IndexTerm><Primary>-ddump-occur-anal option</Primary></IndexTerm>
353 <IndexTerm><Primary>-ddump-spec option</Primary></IndexTerm>
354 <IndexTerm><Primary>-ddump-stg option</Primary></IndexTerm>
355 <IndexTerm><Primary>-ddump-absC option</Primary></IndexTerm>
356 <IndexTerm><Primary>-ddump-flatC option</Primary></IndexTerm>
357 <IndexTerm><Primary>-ddump-realC option</Primary></IndexTerm>
358 <IndexTerm><Primary>-ddump-asm option</Primary></IndexTerm>
363 <Term><Literal>-dverbose-simpl</Literal> and <Literal>-dverbose-stg</Literal>:</Term>
366 <IndexTerm><Primary>-dverbose-simpl option</Primary></IndexTerm>
367 <IndexTerm><Primary>-dverbose-stg option</Primary></IndexTerm>
368 Show the output of the intermediate Core-to-Core and STG-to-STG
369 passes, respectively. (<Emphasis>Lots</Emphasis> of output!) So: when we're
373 % ghc -noC -O -ddump-simpl -dverbose-simpl -dcore-lint Foo.hs
380 <Term><Literal>-ddump-simpl-iterations</Literal>:</Term>
383 <IndexTerm><Primary>-ddump-simpl-iterations option</Primary></IndexTerm>
384 Show the output of each <Emphasis>iteration</Emphasis> of the simplifier (each run of
385 the simplifier has a maximum number of iterations, normally 4). Used
386 when even <Literal>-dverbose-simpl</Literal> doesn't cut it.
391 <Term><Literal>-dppr-{user,debug</Literal>}:</Term>
394 <IndexTerm><Primary>-dppr-user option</Primary></IndexTerm>
395 <IndexTerm><Primary>-dppr-debug option</Primary></IndexTerm>
396 Debugging output is in one of several ``styles.'' Take the printing
397 of types, for example. In the ``user'' style, the compiler's internal
398 ideas about types are presented in Haskell source-level syntax,
399 insofar as possible. In the ``debug'' style (which is the default for
400 debugging output), the types are printed in with
401 explicit foralls, and variables have their unique-id attached (so you
402 can check for things that look the same but aren't).
407 <Term><Literal>-ddump-simpl-stats</Literal>:</Term>
410 <IndexTerm><Primary>-ddump-simpl-stats option</Primary></IndexTerm>
411 Dump statistics about how many of each kind
412 of transformation too place. If you add <Literal>-dppr-debug</Literal> you get more detailed information.
417 <Term><Literal>-ddump-raw-asm</Literal>:</Term>
420 <IndexTerm><Primary>-ddump-raw-asm option</Primary></IndexTerm>
421 Dump out the assembly-language stuff, before the ``mangler'' gets it.
426 <Term><Literal>-ddump-rn-trace</Literal>:</Term>
429 <IndexTerm><Primary>-ddump-rn-trace</Primary></IndexTerm>
430 Make the renamer be *real* chatty about what it is upto.
435 <Term><Literal>-dshow-rn-stats</Literal>:</Term>
438 <IndexTerm><Primary>-dshow-rn-stats</Primary></IndexTerm>
439 Print out summary of what kind of information the renamer had to bring
445 <Term><Literal>-dshow-unused-imports</Literal>:</Term>
448 <IndexTerm><Primary>-dshow-unused-imports</Primary></IndexTerm>
449 Have the renamer report what imports does not contribute.
458 <Sect2 id="checking-consistency">
459 <Title>Checking for consistency
463 <IndexTerm><Primary>consistency checks</Primary></IndexTerm>
464 <IndexTerm><Primary>lint</Primary></IndexTerm>
471 <Term><Literal>-dcore-lint</Literal>:</Term>
474 <IndexTerm><Primary>-dcore-lint option</Primary></IndexTerm>
475 Turn on heavyweight intra-pass sanity-checking within GHC, at Core
476 level. (It checks GHC's sanity, not yours.)
481 <Term><Literal>-dstg-lint</Literal>:</Term>
484 <IndexTerm><Primary>-dstg-lint option</Primary></IndexTerm>
490 <Term><Literal>-dusagesp-lint</Literal>:</Term>
493 <IndexTerm><Primary>-dstg-lint option</Primary></IndexTerm>
494 Turn on checks around UsageSP inference (<Literal>-fusagesp</Literal>). This verifies
495 various simple properties of the results of the inference, and also
496 warns if any identifier with a used-once annotation before the
497 inference has a used-many annotation afterwards; this could indicate a
498 non-worksafe transformation is being applied.
508 <Title>How to read Core syntax (from some <Literal>-ddump-*</Literal> flags)</Title>
511 <IndexTerm><Primary>reading Core syntax</Primary></IndexTerm>
512 <IndexTerm><Primary>Core syntax, how to read</Primary></IndexTerm>
516 Let's do this by commenting an example. It's from doing
517 <Literal>-ddump-ds</Literal> on this code:
520 skip2 m = m : skip2 (m+2)
526 Before we jump in, a word about names of things. Within GHC,
527 variables, type constructors, etc., are identified by their
528 ``Uniques.'' These are of the form `letter' plus `number' (both
529 loosely interpreted). The `letter' gives some idea of where the
530 Unique came from; e.g., <Literal>_</Literal> means ``built-in type variable'';
531 <Literal>t</Literal> means ``from the typechecker''; <Literal>s</Literal> means ``from the
532 simplifier''; and so on. The `number' is printed fairly compactly in
533 a `base-62' format, which everyone hates except me (WDP).
537 Remember, everything has a ``Unique'' and it is usually printed out
538 when debugging, in some form or another. So here we go…
544 Main.skip2{-r1L6-} :: _forall_ a$_4 =>{{Num a$_4}} -> a$_4 -> [a$_4]
546 --# `r1L6' is the Unique for Main.skip2;
547 --# `_4' is the Unique for the type-variable (template) `a'
548 --# `{{Num a$_4}}' is a dictionary argument
552 --# `_NI_' means "no (pragmatic) information" yet; it will later
553 --# evolve into the GHC_PRAGMA info that goes into interface files.
556 /\ _4 -> \ d.Num.t4Gt ->
559 +.t4Hg :: _4 -> _4 -> _4
561 +.t4Hg = (+{-r3JH-} _4) d.Num.t4Gt
563 fromInt.t4GS :: Int{-2i-} -> _4
565 fromInt.t4GS = (fromInt{-r3JX-} _4) d.Num.t4Gt
567 --# The `+' class method (Unique: r3JH) selects the addition code
568 --# from a `Num' dictionary (now an explicit lamba'd argument).
569 --# Because Core is 2nd-order lambda-calculus, type applications
570 --# and lambdas (/\) are explicit. So `+' is first applied to a
571 --# type (`_4'), then to a dictionary, yielding the actual addition
572 --# function that we will use subsequently...
574 --# We play the exact same game with the (non-standard) class method
575 --# `fromInt'. Unsurprisingly, the type `Int' is wired into the
585 } in fromInt.t4GS ds.d4Qz
587 --# `I# 2#' is just the literal Int `2'; it reflects the fact that
588 --# GHC defines `data Int = I# Int#', where Int# is the primitive
589 --# unboxed type. (see relevant info about unboxed types elsewhere...)
591 --# The `!' after `I#' indicates that this is a *saturated*
592 --# application of the `I#' data constructor (i.e., not partially
595 skip2.t3Ja :: _4 -> [_4]
599 let { ds.d4QQ :: [_4]
605 ds.d4QY = +.t4Hg m.r1H4 lit.t4Hb
606 } in skip2.t3Ja ds.d4QY
616 (``It's just a simple functional language'' is an unregisterised
617 trademark of Peyton Jones Enterprises, plc.)
622 <Sect2 id="source-file-options">
623 <Title>Command line options in source files
627 <IndexTerm><Primary>source-file options</Primary></IndexTerm>
631 Sometimes it is useful to make the connection between a source file
632 and the command-line options it requires quite tight. For instance,
633 if a (Glasgow) Haskell source file uses <Literal>casm</Literal>s, the C back-end
634 often needs to be told about which header files to include. Rather than
635 maintaining the list of files the source depends on in a
636 <Literal>Makefile</Literal> (using the <Literal>-#include</Literal> command-line option), it is
637 possible to do this directly in the source file using the <Literal>OPTIONS</Literal>
638 pragma <IndexTerm><Primary>OPTIONS pragma</Primary></IndexTerm>:
643 {-# OPTIONS -#include "foo.h" #-}
651 <Literal>OPTIONS</Literal> pragmas are only looked for at the top of your source
652 files, upto the first (non-literate,non-empty) line not containing
653 <Literal>OPTIONS</Literal>. Multiple <Literal>OPTIONS</Literal> pragmas are recognised. Note
654 that your command shell does not get to the source file options, they
655 are just included literally in the array of command-line arguments
656 the compiler driver maintains internally, so you'll be desperately
657 disappointed if you try to glob etc. inside <Literal>OPTIONS</Literal>.
661 NOTE: the contents of OPTIONS are prepended to the command-line
662 options, so you *do* have the ability to override OPTIONS settings
663 via the command line.
667 It is not recommended to move all the contents of your Makefiles into
668 your source files, but in some circumstances, the <Literal>OPTIONS</Literal> pragma
669 is the Right Thing. (If you use <Literal>-keep-hc-file-too</Literal> and have OPTION
670 flags in your module, the OPTIONS will get put into the generated .hc