make/unit-tests/directive-for.mk

# $NetBSD: directive-for.mk,v 1.26 2024/07/05 19:47:22 rillig Exp $
#
# Tests for the .for directive.
#
# TODO: Describe naming conventions for the loop variables.
#	.for f in values
#	.for file in values
#	.for _FILE_ in values
#	.for .FILE. in values
#	.for _f_ in values
#
# See also:
#	varmod-loop.mk		The ':@var (a] ...@' modifier

# 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}
.endfor
.if ${NUMBERS} != "1 2 3"
.  error
.endif


# The .for loop also works for multiple iteration variables.
# This is something that the modifier :@ cannot do as easily.
.for name value in VARNAME value NAME2 value2
${name}=	${value}
.endfor
.if ${VARNAME} != "value" || ${NAME2} != "value2"
.  error
.endif


# The .for loop splits the items at whitespace, taking quotes into account,
# 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
.endfor
.if ${WORDS:[#]} != 6
.  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, before interpreting the body of the .for loop, the body is
# generated by replacing each expression ${var} with ${:U1}, ${:U2} and so
# on.
#
# 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.  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
.endfor
.if ${var} != "value before"
.  warning After the .for loop, var must still have its original value.
.endif
.if ${var2} != "value before"
.  warning After the .for loop, var2 must still have its original value.
.endif

# Everything from the paragraph above also applies if the loop body is
# 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}
.endfor
.if ${var} != "value before"
.  warning After the .for loop, var must still have its original value.
.endif
.if ${var2} != "value before"
.  warning After the .for loop, var2 must still have its original value.
.endif

# Before for.c 1.39 from 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, such as generating '.if'
# directives or inserting '$' characters in random places, thereby changing
# how following '$' are interpreted.
#
# Before that date, the .for loop below expanded to:
#	EXPANSION+= value
# Since that date, the .for loop below expands to:
#	EXPANSION${:U+}= value
#
EXPANSION=		before
EXPANSION+ =		before
.for plus in +
EXPANSION${plus}=	value
.endfor
.if ${EXPANSION} != "before"
.  error This must be a make from before 2009.
.endif
.if ${EXPANSION+} != "value"
.  error This must be a make from before 2009.
.endif

# When the outer .for loop is expanded, it sees the expression ${i} and
# 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
# value literally in the body.  This could lead to situations where the
# characters from the variable value were interpreted as markup rather than
# plain text.
#
# Until 2012-06-03, the .for loop had split the words at whitespace, without
# taking quotes into account.  This made it possible to have variable values
# 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.
.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.
# See ForLoop_SubstBody.
.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: }{ }{ }{

# 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}>
.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


# Regular global variables and the "variables" from the .for loop don't
# interfere with each other.  In the following snippet, the variable 'DIRECT'
# is used both as a global variable, as well as an iteration variable in the
# .for loop.  The expression '${INDIRECT}' refers to the global variable, not
# to the one from the .for loop.
DIRECT=		global
INDIRECT=	${DIRECT}
.for DIRECT in iteration
.  if "${DIRECT} ${INDIRECT}" != "iteration global"
.    error
.  endif
.endfor


# XXX: A parse error or evaluation error in the items of the .for loop
# should skip the whole loop.  As of 2023-05-09, the loop is expanded as
# usual.
# expect+1: while evaluating "${:Uword2:Z}-after word3" with value "word2": Unknown modifier "Z"
.for var in word1 before-${:Uword2:Z}-after word3
.  info XXX: Should not reach ${var}
.endfor
# expect-2: XXX: Should not reach word1
# expect-3: XXX: Should not reach before--after
# expect-4: XXX: Should not reach 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
.  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.
.for var in
.  error
.endfor

# If the iteration values become empty after expanding the expressions, the
# body of the loop is not evaluated.  It is not a parse error.
.for var in ${:U}
.  error
.endfor


# The loop body can be empty.
.for var in 1 2 3
.endfor


# A mismatched .if inside a .for loop is detected each time when the loop body
# is processed.
.for var in value
.  if 0
.endfor				# expect+0: 1 open conditional

# If there are no iteration values, the loop body is not processed, and the
# check for mismatched conditionals is not performed.
.for var in ${:U}
.  if 0
.endfor


# When a .for without the corresponding .endfor occurs in an inactive branch
# of an .if, the .for directive is just skipped, it does not even need a
# corresponding .endfor.  In other words, the behavior of the parser depends
# on the actual values of the conditions in the .if clauses.
.if 0
.  for var in value		# does not need a corresponding .endfor
.endif
.endfor				# expect+0: for-less endfor
.endif				# expect+0: if-less endif


# When a .for without the corresponding .endfor occurs in an active branch of
# an .if, the parser just counts the number of .for and .endfor directives,
# without looking at any other directives.
.if 1
.  for var in value
.    endif			# expect+0: if-less endif
.  endfor			# no 'for-less endfor'
.endif				# no 'if-less endif'


# 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
.\
   for inner in i
.\
   endfor
.endfor
.MAKEFLAGS: -d0


# When there is a variable definition 'scope=cmdline' from the command line
# (which has higher precedence than global variables) and a .for loop iterates
# over a variable of the same name, the expression '${scope}' expands to the
# value from the .for loop.  This is because when the body of the .for loop is
# expanded, the expression '${scope}' is textually replaced with ${:Uloop}',
# without resolving any other variable names (ForLoop_SubstBody).  Later, when
# the body of the .for loop is actually interpreted, the body text doesn't
# contain the word 'scope' anymore.
.MAKEFLAGS: scope=cmdline
.for scope in loop
.  if ${scope} != "loop"
.    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)