varparse-undef-partial.mk revision 1.3
1 1.3 rillig # $NetBSD: varparse-undef-partial.mk,v 1.3 2020/11/04 05:10:01 rillig Exp $ 2 1.2 rillig 3 1.2 rillig # When an undefined variable is expanded in a ':=' assignment, only the 4 1.2 rillig # initial '$' of the variable expression is skipped by the parser, while 5 1.2 rillig # the remaining expression is evaluated. In edge cases this can lead to 6 1.2 rillig # a completely different interpretation of the partially expanded text. 7 1.2 rillig 8 1.2 rillig LIST= ${DEF} ${UNDEF} ${VAR.${PARAM}} end 9 1.2 rillig DEF= defined 10 1.2 rillig PARAM= :Q 11 1.2 rillig 12 1.3 rillig # The expression ${VAR.${PARAM}} refers to the variable named "VAR.:Q", 13 1.2 rillig # with the ":Q" being part of the name. This variable is not defined, 14 1.2 rillig # therefore the initial '$' of that whole expression is skipped by the 15 1.2 rillig # parser (see Var_Subst, the Buf_AddByte in the else branch) and the rest 16 1.2 rillig # of the expression is expanded as usual. 17 1.2 rillig # 18 1.2 rillig # The resulting variable expression is ${VAR.:Q}, which means that the 19 1.2 rillig # interpretation of the ":Q" has changed from being part of the variable 20 1.2 rillig # name to being a variable modifier. This is a classical code injection. 21 1.2 rillig EVAL:= ${LIST} 22 1.2 rillig .if ${EVAL} != "defined end" 23 1.2 rillig . error ${EVAL} 24 1.2 rillig .endif 25 1.2 rillig 26 1.2 rillig # Define the possible outcomes, to see which of them gets expanded. 27 1.2 rillig VAR.= var-dot without parameter 28 1.2 rillig ${:UVAR.\:Q}= var-dot with parameter :Q 29 1.2 rillig 30 1.2 rillig # At this point, the variable "VAR." is defined, therefore the expression 31 1.3 rillig # ${VAR.:Q} is expanded, consisting of the variable name "VAR." and the 32 1.3 rillig # modifier ":Q". 33 1.2 rillig .if ${EVAL} != "defined var-dot\\ without\\ parameter end" 34 1.2 rillig . error ${EVAL} 35 1.2 rillig .endif 36 1.2 rillig 37 1.2 rillig # In contrast to the previous line, evaluating the original LIST again now 38 1.3 rillig # produces a different result since the variable named "VAR.:Q" is now 39 1.3 rillig # defined. It is expanded as usual, interpreting the ":Q" as part of the 40 1.2 rillig # variable name, as would be expected from reading the variable expression. 41 1.2 rillig EVAL:= ${LIST} 42 1.2 rillig .if ${EVAL} != "defined var-dot with parameter :Q end" 43 1.2 rillig . error ${EVAL} 44 1.2 rillig .endif 45 1.2 rillig 46 1.2 rillig # It's difficult to decide what the best behavior is in this situation. 47 1.2 rillig # Should the whole expression be skipped for now, or should the inner 48 1.2 rillig # subexpressions be expanded already? 49 1.2 rillig # 50 1.2 rillig # Example 1: 51 1.2 rillig # CFLAGS:= ${CFLAGS:N-W*} ${COPTS.${COMPILER}} 52 1.2 rillig # 53 1.2 rillig # The variable COMPILER typically contains an identifier and the variable is 54 1.2 rillig # not modified later. In this practical case, it does not matter whether the 55 1.2 rillig # expression is expanded early, or whether the whole ${COPTS.${COMPILER}} is 56 1.2 rillig # expanded as soon as the variable COPTS.${COMPILER} becomes defined. The 57 1.2 rillig # expression ${COMPILER} would be expanded several times, but in this simple 58 1.2 rillig # scenario there would not be any side effects. 59 1.2 rillig # 60 1.2 rillig # TODO: Add a practical example where early/lazy expansion actually makes a 61 1.2 rillig # difference. 62 1.2 rillig 63 1.2 rillig all: 64 1.2 rillig @: 65
Indexes created Sun Oct 12 09:09:55 GMT 2025