# $NetBSD: dep-var.mk,v 1.5 2020/09/13 20:04:26 rillig Exp $ # # Tests for variable references in dependency declarations. # # Uh oh, this feels so strange that probably nobody uses it. But it seems to # be the only way to reach the lower half of SuffExpandChildren. # XXX: The -dv log says: # Var_Parse: ${UNDEF1} with VARE_UNDEFERR|VARE_WANTRES # but no error message is generated for this line. # The variable expression ${UNDEF1} simply expands to an empty string. all: ${UNDEF1} # Using a double dollar in order to circumvent immediate variable expansion # feels like unintended behavior. At least the manual page says nothing at # all about defined or undefined variables in dependency lines. # # At the point where the expression ${DEF2} is expanded, the variable DEF2 # is defined, so everything's fine. all: $${DEF2} a-$${DEF2}-b # This variable is not defined at all. # XXX: The -dv log says: # Var_Parse: ${UNDEF3} with VARE_UNDEFERR|VARE_WANTRES # but no error message is generated for this line, just like for UNDEF1. # The variable expression ${UNDEF3} simply expands to an empty string. all: $${UNDEF3} # Try out how many levels of indirection are really expanded in dependency # lines. # # The first level of indirection is the $$ in the dependency line. # When the dependency line is parsed, it is resolved to the string # "${INDIRECT_1}". At this point, the dollar is just an ordinary character, # waiting to be expanded at some later point. # # Later, in SuffExpandChildren, that expression is expanded again by calling # Var_Parse, and this time, the result is the string "1-2-${INDIRECT_2}-2-1". # # This string is not expanded anymore by Var_Parse. But there is another # effect. Now DirExpandCurly comes into play and expands the curly braces # in this filename pattern, resulting in the string "1-2-$INDIRECT_2-2-1". # As of 2020-09-03, the test dir.mk contains further details on this topic. # # Finally, this string is assigned to the local ${.TARGET} variable. This # variable is expanded when the shell command is generated. At that point, # the $I is expanded. Since the variable I is not defined, it expands to # the empty string. This way, the final output is the string # "1-2-NDIRECT_2-2-1", which differs from the actual name of the target. # For exactly this reason, it is not recommended to use dollar signs in # target names. # # The number of actual expansions is way more than one might expect, # therefore this feature is probably not widely used. # all: 1-$${INDIRECT_1}-1 INDIRECT_1= 2-$${INDIRECT_2}-2 INDIRECT_2= 3-$${INDIRECT_3}-3 INDIRECT_3= indirect UNDEF1= undef1 DEF2= def2 # Cover the code in SuffExpandChildren that deals with malformed variable # expressions. # # This seems to be an edge case that never happens in practice, and it would # probably be appropriate to just error out in such a case. # # To trigger this piece of code, the variable name must contain "$)" or "$:" # or "$)" or "$$". Using "$:" does not work since the dependency line is # fully expanded before parsing, therefore any ':' in a target or source name # would be interpreted as a dependency operator instead. all: $$$$) # The $$INDIRECT in the following line is treated like the dependency of the # "all" target, that is, the "$$I" is first expanded to "$I", and in a second # round of expansion, the "$I" expands to nothing since the variable "I" is # undefined. # # Since 2020-09-13, this generates a parse error in lint mode (-dL), but not # in normal mode since ParseDoDependency does not handle any errors after # calling Var_Parse. undef1 def2 a-def2-b 1-2-$$INDIRECT_2-2-1 ${:U\$)}: @echo ${.TARGET:Q} # XXX: Why is the exit status still 0, even though Parse_Error is called # with PARSE_FATAL in SuffExpandChildren? |