1 files changed, 105 insertions, 44 deletions

diff --git a/contrib/bmake/unit-tests/directive-for.mk b/contrib/bmake/unit-tests/directive-for.mk
index 95171c68031f..224a466a7709 100755
--- a/contrib/bmake/unit-tests/directive-for.mk
+++ b/contrib/bmake/unit-tests/directive-for.mk
@@ -1,4 +1,4 @@
-# $NetBSD: directive-for.mk,v 1.15 2022/10/01 09:23:04 rillig Exp $
+# $NetBSD: directive-for.mk,v 1.20 2023/05/10 13:03:06 rillig Exp $
 #
 # Tests for the .for directive.
 #
@@ -8,11 +8,15 @@
 #	.for _FILE_ in values
 #	.for .FILE. in values
 #	.for _f_ in values
-
-# Using the .for loop, lists of values can be produced.
-# In simple cases, the :@var@${var}@ variable modifier can be used to
-# achieve the same effects.
 #
+# See also:
+#	varmod-loop.mk		The ':@var@...@' modifier
+
+# expect-all
+
+# A typical use case for a .for loop is to populate a variable with a list of
+# values depending on other variables.  In simple cases, the same effect can
+# be achieved using the ':@var@${var}@' modifier.
 .undef NUMBERS
 .for num in 1 2 3
 NUMBERS+=	${num}
@@ -21,8 +25,9 @@ NUMBERS+=	${num}
 .  error
 .endif
 
+
 # The .for loop also works for multiple iteration variables.
-# This is something that the variable modifier :@ cannot do.
+# This is something that the modifier :@ cannot do.
 .for name value in VARNAME value NAME2 value2
 ${name}=	${value}
 .endfor
@@ -30,12 +35,12 @@ ${name}=	${value}
 .  error
 .endif
 
+
 # The .for loop splits the items at whitespace, taking quotes into account,
-# just like the :M or :S variable modifiers.
-#
-# Until 2012-06-03, it had split the items exactly at whitespace, without
-# taking the quotes into account.  This had resulted in 10 words.
+# just like the :M or :S modifiers.
 #
+# Until 2012-06-03, the .for loop had split the items exactly at whitespace,
+# without taking the quotes into account.  This had resulted in 10 words.
 .undef WORDS
 .for var in one t\ w\ o "three three" 'four four' `five six`
 WORDS+=	counted
@@ -44,16 +49,19 @@ WORDS+=	counted
 .  error
 .endif
 
+
 # In the body of the .for loop, the iteration variables can be accessed
 # like normal variables, even though they are not really variables.
 #
-# Instead, the expression ${var} is transformed into ${:U1}, ${:U2} and so
-# on, before the loop body is evaluated.
+# Instead, before interpreting the body of the .for loop, the body is
+# generated by replacing each expression ${var} with ${:U1}, ${:U2} and so
+# on.
 #
-# A notable effect of this implementation technique is that the .for
+# A noticeable effect of this implementation technique is that the .for
 # iteration variables and the normal global variables live in separate
-# namespaces and do not influence each other.
-#
+# namespaces and do not influence each other.  The "scope" of the .for loop
+# variables is restricted to the current makefile, it does not reach over to
+# any included makefiles.
 var=	value before
 var2=	value before
 .for var var2 in 1 2 3 4
@@ -66,9 +74,8 @@ var2=	value before
 .endif
 
 # Everything from the paragraph above also applies if the loop body is
-# empty, even if there is no actual iteration since the loop items are
-# also empty.
-#
+# empty.  In this particular example, the items to be iterated are empty as
+# well.
 var=	value before
 var2=	value before
 .for var var2 in ${:U}
@@ -82,11 +89,13 @@ var2=	value before
 
 # Until 2008-12-21, the values of the iteration variables were simply
 # inserted as plain text and then parsed as usual, which made it possible
-# to achieve all kinds of strange effects.
+# to achieve all kinds of strange effects, such as generating '.if'
+# directives or inserting '$' characters in random places, thereby changing
+# how following '$' are interpreted.
 #
-# Before that date, the .for loop expanded to:
+# Before that date, the .for loop below expanded to:
 #	EXPANSION+= value
-# Since that date, the .for loop expands to:
+# Since that date, the .for loop below expands to:
 #	EXPANSION${:U+}= value
 #
 EXPANSION=		before
@@ -102,13 +111,16 @@ EXPANSION${plus}=	value
 .endif
 
 # When the outer .for loop is expanded, it sees the expression ${i} and
-# expands it.  The inner loop then has nothing more to expand.
+# expands it.  The inner loop then only sees the expression ${:Uouter} and
+# has nothing more to expand.
 .for i in outer
 .  for i in inner
+# expect+1: outer
 .    info ${i}
 .  endfor
 .endfor
 
+
 # From https://gnats.netbsd.org/29985.
 #
 # Until 2008-12-21, the .for loop was expanded by replacing the variable
@@ -121,17 +133,13 @@ EXPANSION${plus}=	value
 # like "a:\ a:\file.txt" that ended in a single backslash.  Since then, the
 # variable values have been replaced with expressions of the form ${:U...},
 # which are not interpreted as code anymore.
-#
-# As of 2020-09-22, a comment in for.c says that it may be possible to
-# produce an "unwanted substitution", but there is no demonstration code yet.
-#
-# The above changes prevent a backslash at the end of a word from being
-# interpreted as part of the code.  Because of this, the trailingBackslash
-# hack in Var_Subst is no longer needed and as of 2020-09-22, has been
-# removed.
 .for path in a:\ a:\file.txt d:\\ d:\\file.txt
 .  info ${path}
 .endfor
+# expect-2: a:\ a:\file.txt
+# expect-3: d:\\
+# expect-4: d:\\file.txt
+
 
 # Ensure that braces and parentheses are properly escaped by the .for loop.
 # Each line must print the same word 3 times.
@@ -139,28 +147,65 @@ EXPANSION${plus}=	value
 .for v in ( [ { ) ] } (()) [[]] {{}} )( ][ }{
 .  info $v ${v} $(v)
 .endfor
+# expect-02: ( ( (
+# expect-03: [ [ [
+# expect-04: { { {
+# expect-05: ) ) )
+# expect-06: ] ] ]
+# expect-07: } } }
+# expect-08: (()) (()) (())
+# expect-09: [[]] [[]] [[]]
+# expect-10: {{}} {{}} {{}}
+# expect-11: )( )( )(
+# expect-12: ][ ][ ][
+# expect-13: }{ }{ }{
 
-# As of 2020-10-25, the variable names may contain arbitrary characters,
-# except for whitespace.  This allows for creative side effects. Hopefully
-# nobody is misusing this "feature".
+# Before 2023-05-09, the variable names could contain arbitrary characters,
+# except for whitespace, allowing for creative side effects, as usual for
+# arbitrary code injection.
 var=	outer
+# expect+1: invalid character ':' in .for loop variable name
 .for var:Q in value "quoted"
-.  info ${var} ${var:Q} ${var:Q:Q}
+.  info <${var}> <${var:Q}> <${var:Q:Q}>
+.endfor
+
+# Before 2023-05-09, when variable names could contain '$', the short
+# expression '$$' was preserved, the long expressions were substituted.
+# expect+1: invalid character '$' in .for loop variable name
+.for $ in value
+.  info <$$> <${$}> <$($)>
+.endfor
+
+
+# https://gnats.netbsd.org/53146 mentions the idea of using a dynamic
+# variable name in .for loops, based on some other variable.  The .for loops
+# are already tricky enough to understand in detail, even without this
+# possibility, therefore the variable names are restricted to using harmless
+# characters only.
+INDIRECT=	direct
+# expect+1: invalid character '$' in .for loop variable name
+.for $(INDIRECT) in value
+# If the variable name could be chosen dynamically, the iteration variable
+# might have been 'direct', thereby expanding the expression '${direct}'.
+.  info <$(INDIRECT)> <$(direct)> <$($(INDIRECT))>
 .endfor
 
 
 # XXX: A parse error or evaluation error in the items of the .for loop
-# should skip the whole loop.  As of 2020-12-27, the loop is expanded twice.
+# should skip the whole loop.  As of 2023-05-09, the loop is expanded as
+# usual.
+# expect+1: Unknown modifier "Z"
 .for var in word1 ${:Uword2:Z} word3
 .  info XXX: Not reached ${var}
 .endfor
+# expect-2: XXX: Not reached word1
+# expect-3: XXX: Not reached word3
 
 
 # An empty list of variables to the left of the 'in' is a parse error.
 .for in value			# expect+0: no iteration variables in for
-# XXX: The loop body is evaluated once, even with the parse error above.
-.  error			# expect+0: Missing argument for ".error"
-.endfor				# expect+0: for-less endfor
+.  error
+.endfor
 
 # An empty list of iteration values to the right of the 'in' is accepted.
 # Unlike in the shell, it is not a parse error.
@@ -214,12 +259,19 @@ var=	outer
 .endif				# no 'if-less endif'
 
 
-# When make parses a .for loop, it assumes that there is no line break between
-# the '.' and the 'for' or 'endfor', as there is no practical reason to break
-# the line at this point.  When make scans the outer .for loop, it does not
-# recognize the inner directives as such.  When make scans the inner .for
-# loop, it recognizes the '.\n for' but does not recognize the '.\n endfor',
-# as LK_FOR_BODY preserves the backslash-newline sequences.
+# Before for.c 1.172 from 2023-05-08, when make parsed a .for loop, it
+# assumed that there was no line continuation between the '.' and the 'for'
+# or 'endfor', as there is no practical reason to break the line at this
+# point.
+#
+# When make scanned the outer .for loop, it did not recognize the inner .for
+# loop as such and instead treated it as an unknown directive.  The body of
+# the outer .for loop thus ended above the '.endfor'.
+#
+# When make scanned the inner .for loop, it did not recognize the inner
+# .endfor as such, which led to a parse error 'Unexpected end of file in .for
+# loop' from the '.endfor' line, followed by a second parse error 'for-less
+# .endfor' from the '.\\n endfor' line.
 .MAKEFLAGS: -df
 .for outer in o
 .\
@@ -244,3 +296,12 @@ var=	outer
 .    error
 .  endif
 .endfor
+
+
+# Since at least 1993, iteration stops at the first newline.
+# Back then, the .newline variable didn't exist, therefore it was unlikely
+# that a newline ever occurred.
+.for var in a${.newline}b${.newline}c
+.  info newline-item=(${var})
+.endfor
+# expect-2: newline-item=(a)