Merge branch 'Phyx/ghc-gh-18382-fix-heap-corruption' into wip/winio-by-defaultwip/winio-by-default

3 files changed, 97 insertions, 19 deletions

diff --git a/docs/users_guide/9.4.1-notes.rst b/docs/users_guide/9.4.1-notes.rst
index 31f09fa6bc..5cb2736547 100644
--- a/docs/users_guide/9.4.1-notes.rst
+++ b/docs/users_guide/9.4.1-notes.rst
@@ -38,6 +38,33 @@ Compiler
     enabled the :extension:`UnboxedTuples` extension but didn't explicitly
     enable :extension:`UnboxedSums` will continue to work without changes.
 
+- Constructed Product Result analysis (c.f. :ghc-flag:`-fcpr-anal`) has been
+  overhauled and will now unbox nestedly, if termination properties of the
+  function permit. This allows unboxing of constructed results returned by
+  ``IO`` actions. E.g.::
+
+      sumIO :: [Int] -> IO Int
+      sumIO []     = return 0
+      sumIO (x:xs) = do
+        r <- sumIO xs
+        return $! x + r
+
+  Note the use of ``$!``: Without it, GHC would be unable to see that evaluation
+  of ``r`` and ``x`` terminates (and rapidly, at that). An alternative would be to
+  evaluate both with a bang pattern or a ``seq``, but the ``return $! <res>``
+  idiom should work more reliably and needs less thinking.
+
+- Demand analysis (cf. :ghc-flag:`-fstrictness`) now integrates a
+  Boxity Analysis that tracks whether a function needs a parameter boxed. If
+  that is the case, the worker/wrapper transformation (cf.
+  :ghc-flag:`-fworker-wrapper`) will not unbox that parameter, leading to less
+  reboxing in many cases.
+
+  For reasons of backwards-compatible performance, you may find that the new
+  mechanism is too aggressive in a few cases (e.g., still unboxing a parameter
+  that is used boxed in a hot path). Do post a bug report with your example!
+  Then wrap the uses of the parameter in ``GHC.Exts.lazy`` for a short-term fix.
+
 ``base`` library
 ~~~~~~~~~~~~~~~~
 
diff --git a/docs/users_guide/editing-guide.rst b/docs/users_guide/editing-guide.rst
index 737abd3b2e..b4720e1a71 100644
--- a/docs/users_guide/editing-guide.rst
+++ b/docs/users_guide/editing-guide.rst
@@ -6,15 +6,15 @@ for the Glasgow Haskell Compiler. Even more than this, it at times serves (for
 better or for worse) as a de-facto language standard, being the sole
 non-academic reference for many widely used language extensions.
 
-Since GHC 8.0, the User's Guide is authored in `ReStructuredText
-<https://en.wikipedia.org/wiki/ReStructuredText>`__ (or ReST or RST, for short)
+Since GHC 8.0, the User's Guide is authored in `reStructuredText
+<https://en.wikipedia.org/wiki/ReStructuredText>`__ (or reST or RST, for short)
 a rich but light-weight mark-up language aimed at producing documentation. The
-`Sphinx <http://sphinx-doc.org/>`__ tool is used to produce the final PDF and
+`Sphinx <https://www.sphinx-doc.org/>`__ tool is used to produce the final PDF and
 HTML documentation.
 
-This document (also written in ReST) serves as a brief introduction to ReST and to
+This document (also written in reST) serves as a brief introduction to reST and to
 document the conventions used in the User's Guide. This document is *not* intended
-to be a thorough guide to ReST. For this see the resources referenced
+to be a thorough guide to reST. For this see the resources referenced
 `below <#references>`__.
 
 Basics
@@ -91,7 +91,7 @@ The above would be rendered as,
 Headings
 ~~~~~~~~
 
-While ReST can accommodate a wide range of heading styles, we have standardized
+While reST can accommodate a wide range of heading styles, we have standardized
 on this convention in the User's Guide,
 
 .. code-block:: rest
@@ -141,7 +141,7 @@ Other languages
 Double-colon blocks are syntax-highlighted as Haskell by default. To avoid this
 use a
 ``.. code-block`` `directive
-<http://sphinx-doc.org/markup/code.html#directive-code-block>`__ with explicit
+<https://www.sphinx-doc.org/en/master/usage/restructuredtext/directives.html#directive-code-block>`__ with explicit
 language designation,
 
 .. code-block:: rest
@@ -314,7 +314,7 @@ There are a number of admonitions types,
     * warning
 
 
-.. _Admonitions: http://docutils.sourceforge.net/docs/ref/rst/directives.html#admonitions
+.. _Admonitions: https://docutils.sourceforge.io/docs/ref/rst/directives.html#admonitions
 
 Documenting command-line options and GHCi commands
 --------------------------------------------------
@@ -420,16 +420,17 @@ user-substitutable tokens. In this document we use the convention, ``⟨subst⟩
 (note that these are angle brackets, ``U+27E8`` and ``U+27E9``, not
 less-than/greater-than signs).
 
+.. _references:
 
-ReST reference materials
+reST reference materials
 ------------------------
 
-* `Sphinx ReST Primer`_: A great place to start.
-* `Sphinx extensions`_: How Sphinx extends ReST
-* `ReST reference`_: When you really need the details.
+* `Sphinx reST Primer`_: A great place to start.
+* `Sphinx extensions`_: How Sphinx extends reST
+* `reST reference`_: When you really need the details.
 * `Directives reference`_
 
-.. _Sphinx ReST Primer: http://sphinx-doc.org/rest.html
-.. _ReST reference: http://docutils.sourceforge.net/docs/ref/rst/restructuredtext.html
-.. _Sphinx extensions: http://sphinx-doc.org/markup/index.html
-.. _Directives reference: http://docutils.sourceforge.net/docs/ref/rst/directives.html#code
+.. _Sphinx reST Primer: https://www.sphinx-doc.org/en/master/usage/restructuredtext/basics.html
+.. _reST reference: https://docutils.sourceforge.io/docs/ref/rst/restructuredtext.html
+.. _Sphinx extensions: https://www.sphinx-doc.org/en/master/usage/extensions/index.html
+.. _Directives reference: https://docutils.sourceforge.io/docs/ref/rst/directives.html#code
diff --git a/docs/users_guide/using-optimisation.rst b/docs/users_guide/using-optimisation.rst
index 37e4a4aae6..139031ad1c 100644
--- a/docs/users_guide/using-optimisation.rst
+++ b/docs/users_guide/using-optimisation.rst
@@ -247,7 +247,7 @@ by saying ``-fno-wombat``.
     :default: on
 
     Enables some control flow optimisations in the Cmm code
-    generator, merging basic blocks and avoiding jumps right after jumps. 
+    generator, merging basic blocks and avoiding jumps right after jumps.
 
 .. ghc-flag:: -fasm-shortcutting
     :shortdesc: Enable shortcutting on assembly. Implied by :ghc-flag:`-O2`.
@@ -319,14 +319,64 @@ by saying ``-fno-wombat``.
     block layout behaves the same as in 8.6 and earlier.
 
 .. ghc-flag:: -fcpr-anal
-    :shortdesc: Turn on CPR analysis in the demand analyser. Implied by :ghc-flag:`-O`.
+    :shortdesc: Turn on Constructed Product Result analysis. Implied by :ghc-flag:`-O`.
     :type: dynamic
     :reverse: -fno-cpr-anal
     :category:
 
     :default: on
 
-    Turn on CPR analysis in the demand analyser.
+    Turn on CPR analysis, which enables the worker/wrapper transformation (cf.
+    :ghc-flag:`-fworker-wrapper`) to unbox the result of a function, such as ::
+
+         sum :: [Int] -> Int
+         sum []     = 0
+         sum (x:xs) = x + sum xs
+
+    CPR analysis will see that each code path produces a *constructed product*
+    such as ``I# 0#`` in the first branch (where ``GHC.Exts.I#`` is the data
+    constructor of ``Int``, boxing up the the primitive integer literal ``0#``
+    of type ``Int#``) and optimise to ::
+
+         sum xs = I# ($wsum xs)
+         $wsum []        = 0#
+         $wsum (I# x:xs) = x# +# $wsum xs
+
+    and then ``sum`` can inline to potentially cancel away the ``I#`` box.
+
+    Here's an example of the function that *does not* return a constructed product: ::
+
+         f :: [Int] -> (Int -> Int) -> Int
+         f []     g = g 0
+         f (x:xs) g = x + f xs g
+
+    The expression ``g 0`` is not a constructed product, because we don't know
+    anything about ``g``.
+
+    CPR analysis also works *nestedly*, for example ::
+
+        sumIO :: [Int] -> IO Int
+        sumIO []     = return 0
+        sumIO (x:xs) = do
+          r <- sumIO xs
+          return $! x + r
+
+    Note the use of ``$!``: Without it, GHC would be unable to see that evaluation
+    of ``r`` and ``x`` terminates (and rapidly, at that). An alternative would be to
+    evaluate both with a bang pattern or a ``seq``, but the ``return $! <res>``
+    idiom should work more reliably and needs less thinking. The above example
+    will be optimised to ::
+
+        sumIO :: [Int] -> IO Int
+        sumIO xs = IO $ \s -> case $wsum xs s of
+          (# s', r #) -> (# s', I# r #)
+        $wsumIO :: [Int] -> (# RealWorld#, Int# #)
+        $wsumIO []        s = (# s, 0# #)
+        $wsumIO (I# x:xs) s = case $wsumIO xs of
+          (# s', r #) -> (# s', x +# r#)
+
+    And the latter can inline ``sumIO`` and cancel away the ``I#`` constructor.
+    Unboxing the result of a ``State`` action should work similarly.
 
 .. ghc-flag:: -fcse
     :shortdesc: Enable common sub-expression elimination. Implied by :ghc-flag:`-O`.