1 <?xml version="1.0" encoding="iso-8859-1"?>
2 <sect1 id="options-debugging">
3 <title>Debugging the compiler</title>
5 <indexterm><primary>debugging options (for GHC)</primary></indexterm>
7 <para>HACKER TERRITORY. HACKER TERRITORY. (You were warned.)</para>
9 <sect2 id="dumping-output">
10 <title>Dumping out compiler intermediate structures</title>
12 <indexterm><primary>dumping GHC intermediates</primary></indexterm>
13 <indexterm><primary>intermediate passes, output</primary></indexterm>
18 <option>-ddump-</option><replaceable>pass</replaceable>
19 <indexterm><primary><option>-ddump</option> options</primary></indexterm>
22 <para>Make a debugging dump after pass
23 <literal><pass></literal> (may be common enough to need
24 a short form…). You can get all of these at once
25 (<emphasis>lots</emphasis> of output) by using
26 <option>-v5</option>, or most of them with
27 <option>-v4</option>. Some of the most useful ones
33 <option>-ddump-parsed</option>:
34 <indexterm><primary><option>-ddump-parsed</option></primary></indexterm>
37 <para>parser output</para>
43 <option>-ddump-rn</option>:
44 <indexterm><primary><option>-ddump-rn</option></primary></indexterm>
47 <para>renamer output</para>
53 <option>-ddump-tc</option>:
54 <indexterm><primary><option>-ddump-tc</option></primary></indexterm>
57 <para>typechecker output</para>
63 <option>-ddump-types</option>:
64 <indexterm><primary><option>-ddump-types</option></primary></indexterm>
67 <para>Dump a type signature for each value defined at
68 the top level of the module. The list is sorted
69 alphabetically. Using <option>-dppr-debug</option>
70 dumps a type signature for all the imported and
71 system-defined things as well; useful for debugging the
78 <option>-ddump-deriv</option>:
79 <indexterm><primary><option>-ddump-deriv</option></primary></indexterm>
82 <para>derived instances</para>
88 <option>-ddump-ds</option>:
89 <indexterm><primary><option>-ddump-ds</option></primary></indexterm>
92 <para>desugarer output</para>
98 <option>-ddump-spec</option>:
99 <indexterm><primary><option>-ddump-spec</option></primary></indexterm>
102 <para>output of specialisation pass</para>
108 <option>-ddump-rules</option>:
109 <indexterm><primary><option>-ddump-rules</option></primary></indexterm>
112 <para>dumps all rewrite rules (including those generated
113 by the specialisation pass)</para>
119 <option>-ddump-simpl</option>:
120 <indexterm><primary><option>-ddump-simpl</option></primary></indexterm>
123 <para>simplifer output (Core-to-Core passes)</para>
129 <option>-ddump-inlinings</option>:
130 <indexterm><primary><option>-ddump-inlinings</option></primary></indexterm>
133 <para>inlining info from the simplifier</para>
139 <option>-ddump-usagesp</option>:
140 <indexterm><primary><option>-ddump-usagesp</option></primary></indexterm>
143 <para>UsageSP inference pre-inf and output</para>
149 <option>-ddump-cpranal</option>:
150 <indexterm><primary><option>-ddump-cpranal</option></primary></indexterm>
153 <para>CPR analyser output</para>
159 <option>-ddump-stranal</option>:
160 <indexterm><primary><option>-ddump-stranal</option></primary></indexterm>
163 <para>strictness analyser output</para>
169 <option>-ddump-cse</option>:
170 <indexterm><primary><option>-ddump-cse</option></primary></indexterm>
173 <para>CSE pass output</para>
179 <option>-ddump-workwrap</option>:
180 <indexterm><primary><option>-ddump-workwrap</option></primary></indexterm>
183 <para>worker/wrapper split output</para>
189 <option>-ddump-occur-anal</option>:
190 <indexterm><primary><option>-ddump-occur-anal</option></primary></indexterm>
193 <para>`occurrence analysis' output</para>
199 <option>-ddump-sat</option>:
200 <indexterm><primary><option>-ddump-sat</option></primary></indexterm>
203 <para>output of “saturate” pass</para>
209 <option>-ddump-stg</option>:
210 <indexterm><primary><option>-ddump-stg</option></primary></indexterm>
213 <para>output of STG-to-STG passes</para>
219 <option>-ddump-absC</option>:
220 <indexterm><primary><option>-ddump-absC</option></primary></indexterm>
223 <para><emphasis>un</emphasis>flattened Abstract C</para>
229 <option>-ddump-flatC</option>:
230 <indexterm><primary><option>-ddump-flatC</option></primary></indexterm>
233 <para><emphasis>flattened</emphasis> Abstract C</para>
239 <option>-ddump-realC</option>:
240 <indexterm><primary><option>-ddump-realC</option></primary></indexterm>
243 <para>same as what goes to the C compiler</para>
249 <option>-ddump-stix</option>:
250 <indexterm><primary><option>-ddump-stix</option></primary></indexterm>
253 <para>native-code generator intermediate form</para>
259 <option>-ddump-asm</option>:
260 <indexterm><primary><option>-ddump-asm</option></primary></indexterm>
263 <para>assembly language from the native-code generator</para>
269 <option>-ddump-bcos</option>:
270 <indexterm><primary><option>-ddump-bcos</option></primary></indexterm>
273 <para>byte code compiler output</para>
279 <option>-ddump-foreign</option>:
280 <indexterm><primary><option>-ddump-foreign</option></primary></indexterm>
283 <para>dump foreign export stubs</para>
292 <option>-dverbose-core2core</option>
293 <indexterm><primary><option>-dverbose-core2core</option></primary></indexterm>
296 <option>-dverbose-stg2stg</option>
297 <indexterm><primary><option>-dverbose-stg2stg</option></primary></indexterm>
300 <para>Show the output of the intermediate Core-to-Core and
301 STG-to-STG passes, respectively. (<emphasis>Lots</emphasis>
302 of output!) So: when we're really desperate:</para>
305 % ghc -noC -O -ddump-simpl -dverbose-simpl -dcore-lint Foo.hs
313 <option>-ddump-simpl-iterations</option>:
314 <indexterm><primary><option>-ddump-simpl-iterations</option></primary></indexterm>
317 <para>Show the output of each <emphasis>iteration</emphasis>
318 of the simplifier (each run of the simplifier has a maximum
319 number of iterations, normally 4). Used when even
320 <option>-dverbose-simpl</option> doesn't cut it.</para>
326 <option>-dppr-debug</option>
327 <indexterm><primary><option>-dppr-debug</option></primary></indexterm>
330 <para>Debugging output is in one of several
331 “styles.” Take the printing of types, for
332 example. In the “user” style (the default), the
333 compiler's internal ideas about types are presented in
334 Haskell source-level syntax, insofar as possible. In the
335 “debug” style (which is the default for
336 debugging output), the types are printed in with explicit
337 foralls, and variables have their unique-id attached (so you
338 can check for things that look the same but aren't). This
339 flag makes debugging output appear in the more verbose debug
346 <option>-dppr-user-length</option>
347 <indexterm><primary><option>-dppr-user-length</option></primary></indexterm>
350 <para>In error messages, expressions are printed to a
351 certain “depth”, with subexpressions beyond the
352 depth replaced by ellipses. This flag sets the
359 <option>-ddump-simpl-stats</option>
360 <indexterm><primary><option>-ddump-simpl-stats option</option></primary></indexterm>
363 <para>Dump statistics about how many of each kind of
364 transformation too place. If you add
365 <option>-dppr-debug</option> you get more detailed
372 <option>-ddump-rn-trace</option>
373 <indexterm><primary><option>-ddump-rn-trace</option></primary></indexterm>
376 <para>Make the renamer be *real* chatty about what it is
383 <option>-ddump-rn-stats</option>
384 <indexterm><primary><option>-dshow-rn-stats</option></primary></indexterm>
387 <para>Print out summary of what kind of information the renamer
388 had to bring in.</para>
394 <option>-dshow-unused-imports</option>
395 <indexterm><primary><option>-dshow-unused-imports</option></primary></indexterm>
398 <para>Have the renamer report what imports does not
405 <sect2 id="checking-consistency">
406 <title>Checking for consistency</title>
408 <indexterm><primary>consistency checks</primary></indexterm>
409 <indexterm><primary>lint</primary></indexterm>
415 <option>-dcore-lint</option>
416 <indexterm><primary><option>-dcore-lint</option></primary></indexterm>
419 <para>Turn on heavyweight intra-pass sanity-checking within
420 GHC, at Core level. (It checks GHC's sanity, not yours.)</para>
426 <option>-dstg-lint</option>:
427 <indexterm><primary><option>-dstg-lint</option></primary></indexterm>
430 <para>Ditto for STG level. (NOTE: currently doesn't work).</para>
436 <option>-dusagesp-lint</option>:
437 <indexterm><primary><option>-dstg-lint</option></primary></indexterm>
440 <para>Turn on checks around UsageSP inference
441 (<option>-fusagesp</option>). This verifies various simple
442 properties of the results of the inference, and also warns
443 if any identifier with a used-once annotation before the
444 inference has a used-many annotation afterwards; this could
445 indicate a non-worksafe transformation is being
453 <title>How to read Core syntax (from some <option>-ddump</option>
456 <indexterm><primary>reading Core syntax</primary></indexterm>
457 <indexterm><primary>Core syntax, how to read</primary></indexterm>
459 <para>Let's do this by commenting an example. It's from doing
460 <option>-ddump-ds</option> on this code:
463 skip2 m = m : skip2 (m+2)
466 Before we jump in, a word about names of things. Within GHC,
467 variables, type constructors, etc., are identified by their
468 “Uniques.” These are of the form `letter' plus
469 `number' (both loosely interpreted). The `letter' gives some idea
470 of where the Unique came from; e.g., <literal>_</literal>
471 means “built-in type variable”; <literal>t</literal>
472 means “from the typechecker”; <literal>s</literal>
473 means “from the simplifier”; and so on. The `number'
474 is printed fairly compactly in a `base-62' format, which everyone
475 hates except me (WDP).</para>
477 <para>Remember, everything has a “Unique” and it is
478 usually printed out when debugging, in some form or another. So
479 here we go…</para>
483 Main.skip2{-r1L6-} :: _forall_ a$_4 =>{{Num a$_4}} -> a$_4 -> [a$_4]
485 --# `r1L6' is the Unique for Main.skip2;
486 --# `_4' is the Unique for the type-variable (template) `a'
487 --# `{{Num a$_4}}' is a dictionary argument
491 --# `_NI_' means "no (pragmatic) information" yet; it will later
492 --# evolve into the GHC_PRAGMA info that goes into interface files.
495 /\ _4 -> \ d.Num.t4Gt ->
498 +.t4Hg :: _4 -> _4 -> _4
500 +.t4Hg = (+{-r3JH-} _4) d.Num.t4Gt
502 fromInt.t4GS :: Int{-2i-} -> _4
504 fromInt.t4GS = (fromInt{-r3JX-} _4) d.Num.t4Gt
506 --# The `+' class method (Unique: r3JH) selects the addition code
507 --# from a `Num' dictionary (now an explicit lamba'd argument).
508 --# Because Core is 2nd-order lambda-calculus, type applications
509 --# and lambdas (/\) are explicit. So `+' is first applied to a
510 --# type (`_4'), then to a dictionary, yielding the actual addition
511 --# function that we will use subsequently...
513 --# We play the exact same game with the (non-standard) class method
514 --# `fromInt'. Unsurprisingly, the type `Int' is wired into the
524 } in fromInt.t4GS ds.d4Qz
526 --# `I# 2#' is just the literal Int `2'; it reflects the fact that
527 --# GHC defines `data Int = I# Int#', where Int# is the primitive
528 --# unboxed type. (see relevant info about unboxed types elsewhere...)
530 --# The `!' after `I#' indicates that this is a *saturated*
531 --# application of the `I#' data constructor (i.e., not partially
534 skip2.t3Ja :: _4 -> [_4]
538 let { ds.d4QQ :: [_4]
544 ds.d4QY = +.t4Hg m.r1H4 lit.t4Hb
545 } in skip2.t3Ja ds.d4QY
553 <para>(“It's just a simple functional language” is an
554 unregisterised trademark of Peyton Jones Enterprises, plc.)</para>
559 <title>Unregisterised compilation</title>
560 <indexterm><primary>unregisterised compilation</primary></indexterm>
562 <para>The term "unregisterised" really means "compile via vanilla
563 C", disabling some of the platform-specific tricks that GHC
564 normally uses to make programs go faster. When compiling
565 unregisterised, GHC simply generates a C file which is compiled
568 <para>Unregisterised compilation can be useful when porting GHC to
569 a new machine, since it reduces the prerequisite tools to
570 <command>gcc</command>, <command>as</command>, and
571 <command>ld</command> and nothing more, and furthermore the amount
572 of platform-specific code that needs to be written in order to get
573 unregisterised compilation going is usually fairly small.</para>
578 <option>-unreg</option>:
579 <indexterm><primary><option>-unreg</option></primary></indexterm>
582 <para>Compile via vanilla ANSI C only, turning off
583 platform-specific optimisations. NOTE: in order to use
584 <option>-unreg</option>, you need to have a set of libraries
585 (including the RTS) built for unregisterised compilation.
586 This amounts to building GHC with way "u" enabled.</para>
595 ;;; Local Variables: ***
597 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter" "sect1") ***