1# $NetBSD: directive-for-empty.mk,v 1.2 2023/06/01 20:56:35 rillig Exp $ 2# 3# Tests for .for loops containing conditions of the form 'empty(var:...)'. 4# 5# When a .for loop is expanded, variable expressions in the body of the loop 6# are replaced with expressions containing the variable values. This 7# replacement is a bit naive but covers most of the practical cases. The one 8# popular exception is the condition 'empty(var:Modifiers)', which does not 9# look like a variable expression and is thus not replaced. 10# 11# See also: 12# https://gnats.netbsd.org/43821 13 14 15# In the body of the .for loop, the expression '${i:M*2*}' is replaced with 16# '${:U11:M*2*}', '${:U12:M*2*}', '${:U13:M*2*}', one after another. This 17# replacement creates the impression that .for variables were real variables, 18# when in fact they aren't. 19.for i in 11 12 13 20. if ${i:M*2*} 21# expect+1: 2 22.info 2 23. endif 24.endfor 25 26 27# In conditions, the function call to 'empty' does not look like a variable 28# expression, therefore it is not replaced. Since there is no global variable 29# named 'i', this expression makes for a leaky abstraction. If the .for 30# variables were real variables, calling 'empty' would work on them as well. 31.for i in 11 12 13 32# Asking for an empty iteration variable does not make sense as the .for loop 33# splits the iteration items into words, and such a word cannot be empty. 34. if empty(i) 35# expect+3: Missing argument for ".error" 36# expect+2: Missing argument for ".error" 37# expect+1: Missing argument for ".error" 38. error # due to the leaky abstraction 39. endif 40# The typical way of using 'empty' with variables from .for loops is pattern 41# matching using the modifiers ':M' or ':N'. 42. if !empty(i:M*2*) 43. if ${i} != "12" 44. error 45. endif 46. endif 47.endfor 48 49 50# The idea of replacing every occurrences of 'empty(i' in the body of a .for 51# loop would be naive and require many special cases, as there are many cases 52# that need to be considered when deciding whether the token 'empty' is a 53# function call or not, as demonstrated by the following examples. For 54# variable expressions like '${i:Modifiers}', this is simpler as a single 55# dollar almost always starts a variable expression. For counterexamples and 56# edge cases, see directive-for-escape.mk. Adding another such tricky detail 57# is out of the question. 58.MAKEFLAGS: -df 59.for i in value 60# The identifier 'empty' can only be used in conditions such as .if, .ifdef or 61# .elif. In other lines the string 'empty(' must be preserved. 62CPPFLAGS+= -Dmessage="empty(i)" 63# There may be whitespace between 'empty' and '('. 64.if ! empty (i) 65. error 66.endif 67# Even in conditions, the string 'empty(' is not always a function call, it 68# can occur in a string literal as well. 69.if "empty\(i)" != "empty(i)" 70. error 71.endif 72# In comments like 'empty(i)', the text must be preserved as well. 73# 74# Conditions, including function calls to 'empty', can not only occur in 75# condition directives, they can also occur in the modifier ':?', see 76# varmod-ifelse.mk. 77CPPFLAGS+= -Dmacro="${empty(i):?empty:not-empty}" 78.endfor 79.MAKEFLAGS: -d0 80 81 82# An idea to work around the above problems is to collect the variables from 83# the .for loops in a separate scope. To match the current behavior, there 84# has to be one scope per included file. There may be .for loops using the 85# same variable name in files that include each other: 86# 87# outer.mk: .for i in outer 88# . info $i # outer 89# . include "inner.mk" 90# inner.mk: . info $i # (undefined) 91# . for i in inner 92# . info $i # inner 93# . endfor 94# . info $i # (undefined) 95# outer.mk: . info $i # outer 96# .endfor 97# 98# This might be regarded another leaky abstraction, but it is in fact useful 99# that variables from .for loops can only affect expressions in the current 100# file. If variables from .for loops were implemented as global variables, 101# they might interact between files. 102# 103# To emulate this exact behavior for the function 'empty', each file in the 104# stack of included files needs its own scope that is independent from the 105# other files. 106# 107# Another tricky detail are nested .for loops in a single file that use the 108# same variable name. These are generally avoided by developers, as they 109# would be difficult to understand for humans as well. Technically, they are 110# possible though. Assuming there are two nested .for loops, both using the 111# variable 'i'. When the inner .for loop ends, the inner 'i' needs to be 112# removed from the scope, which would need to make the outer 'i' visible 113# again. This would suggest to use one variable scope per .for loop. 114# 115# Using a separate scope has the benefit that Var_Parse already allows for 116# a custom scope to be passed as parameter. This would have another side 117# effect though. There are several modifiers that actually modify variables, 118# and these modifications happen in the scope that is passed to Var_Parse. 119# This would mean that the combination of a .for variable and the modifiers 120# '::=', '::+=', '::?=', '::!=' and ':_' would lead to different behavior than 121# before. 122 123# TODO: Add code that demonstrates the current interaction between variables 124# from .for loops and the modifiers mentioned above. 125