1# $NetBSD: directive-for-escape.mk,v 1.12 2021/12/05 11:40:03 rillig Exp $ 2# 3# Test escaping of special characters in the iteration values of a .for loop. 4# These values get expanded later using the :U variable modifier, and this 5# escaping and unescaping must pass all characters and strings effectively 6# unmodified. 7 8.MAKEFLAGS: -df 9 10# Even though the .for loops take quotes into account when splitting the 11# string into words, the quotes don't need to be balanced, as of 2020-12-31. 12# This could be considered a bug. 13ASCII= !"\#$$%&'()*+,-./0-9:;<=>?@A-Z[\]_^a-z{|}~ 14 15# XXX: As of 2020-12-31, the '#' is not preserved in the expanded body of 16# the loop. Not only would it need the escaping for the variable modifier 17# ':U' but also the escaping for the line-end comment. 18.for chars in ${ASCII} 19. info ${chars} 20.endfor 21 22# As of 2020-12-31, using 2 backslashes before be '#' would treat the '#' 23# as comment character. Using 3 backslashes doesn't help either since 24# then the situation is essentially the same as with 1 backslash. 25# This means that a '#' sign cannot be passed in the value of a .for loop 26# at all. 27ASCII.2020-12-31= !"\\\#$$%&'()*+,-./0-9:;<=>?@A-Z[\]_^a-z{|}~ 28.for chars in ${ASCII.2020-12-31} 29. info ${chars} 30.endfor 31 32# Cover the code in ExprLen. 33# 34# XXX: It is unexpected that the variable V gets expanded in the loop body. 35# The double '$$' should intuitively prevent exactly this. Probably nobody 36# was adventurous enough to use literal dollar signs in the values of a .for 37# loop, allowing this edge case to go unnoticed for years. 38# 39# See for.c, function ExprLen. 40V= value 41VALUES= $$ $${V} $${V:=-with-modifier} $$(V) $$(V:=-with-modifier) 42.for i in ${VALUES} 43. info $i 44.endfor 45 46# Try to cover the code for nested '{}' in ExprLen, without success. 47# 48# The value of the variable VALUES is not meant to be a variable expression. 49# Instead, it is meant to represent literal text, the only escaping mechanism 50# being that each '$' is written as '$$'. 51# 52# The .for loop splits ${VALUES} into 3 words, at the space characters, since 53# these are not escaped. 54VALUES= $${UNDEF:U\$$\$$ {{}} end} 55# XXX: Where in the code does the '\$\$' get converted into a single '\$'? 56.for i in ${VALUES} 57. info $i 58.endfor 59 60# Second try to cover the code for nested '{}' in ExprLen. 61# 62# XXX: It is wrong that ExprLen requires the braces to be balanced. 63# Each variable modifier has its own inconsistent way of parsing nested 64# variable expressions, braces and parentheses. (Compare ':M', ':S', and 65# ':D' for details.) The only sensible thing to do is therefore to let 66# Var_Parse do all the parsing work. 67VALUES= begin<$${UNDEF:Ufallback:N{{{}}}}>end 68.for i in ${VALUES} 69. info $i 70.endfor 71 72# A single trailing dollar doesn't happen in practice. 73# The dollar sign is correctly passed through to the body of the .for loop. 74# There, it is expanded by the .info directive, but even there a trailing 75# dollar sign is kept as-is. 76.for i in ${:U\$} 77. info ${i} 78.endfor 79 80# As of 2020-12-31, the name of the iteration variable can even contain 81# colons, which then affects variable expressions having this exact modifier. 82# This is clearly an unintended side effect of the implementation. 83NUMBERS= one two three 84.for NUMBERS:M*e in replaced 85. info ${NUMBERS} ${NUMBERS:M*e} 86.endfor 87 88# As of 2020-12-31, the name of the iteration variable can contain braces, 89# which gets even more surprising than colons, since it allows to replace 90# sequences of variable expressions. There is no practical use case for 91# this, though. 92BASENAME= one 93EXT= .c 94.for BASENAME}${EXT in replaced 95. info ${BASENAME}${EXT} 96.endfor 97 98# Demonstrate the various ways to refer to the iteration variable. 99i= outer 100i2= two 101i,= comma 102.for i in inner 103. info . $$i: $i 104. info . $${i}: ${i} 105. info . $${i:M*}: ${i:M*} 106. info . $$(i): $(i) 107. info . $$(i:M*): $(i:M*) 108. info . $${i$${:U}}: ${i${:U}} 109. info . $${i\}}: ${i\}} # XXX: unclear why ForLoop_SubstVarLong needs this 110. info . $${i2}: ${i2} 111. info . $${i,}: ${i,} 112. info . adjacent: $i${i}${i:M*}$i 113.endfor 114 115# The variable name can be a single '$' since there is no check on valid 116# variable names. ForLoop_SubstVarShort skips "stupid" variable names though, 117# but ForLoop_SubstVarLong naively parses the body of the loop, substituting 118# each '${$}' with an actual 'dollar'. 119.for $ in dollar 120. info eight $$$$$$$$ and no cents. 121. info eight ${$}${$}${$}${$} and no cents. 122.endfor 123# Outside a .for loop, '${$}' is interpreted differently. The outer '$' starts 124# a variable expression. The inner '$' is followed by a '}' and is thus a 125# silent syntax error, the '$' is skipped. The variable name is thus '', and 126# since since there is never a variable named '', the whole expression '${$}' 127# evaluates to an empty string. 128closing-brace= } # guard against an 129${closing-brace}= <closing-brace> # alternative interpretation 130.info eight ${$}${$}${$}${$} and no cents. 131 132# What happens if the values from the .for loop contain a literal newline? 133# Before for.c 1.144 from 2021-06-25, the newline was passed verbatim to the 134# body of the .for loop, where it was then interpreted as a literal newline, 135# leading to syntax errors such as "Unclosed variable expression" in the upper 136# line and "Invalid line type" in the lower line. 137.for i in "${.newline}" 138. info short: $i 139. info long: ${i} 140.endfor 141 142all: 143