# $NetBSD: opt-debug-lint.mk,v 1.15 2023/06/01 20:56:35 rillig Exp $ # # Tests for the -dL command line option, which runs additional checks # to catch common mistakes, such as unclosed variable expressions. .MAKEFLAGS: -dL # Since 2020-09-13, undefined variables that are used on the left-hand side # of a condition at parse time get a proper error message. Before, the # error message was "Malformed conditional" only, which was wrong and # misleading. The form of the condition is totally fine, it's the evaluation # that fails. # # Since 2020-09-13, the "Malformed conditional" error message is not printed # anymore. # # See also: # cond-undef-lint.mk # expect+2: Malformed conditional ($X) # expect+1: Variable "X" is undefined .if $X . error .endif # The dynamic variables like .TARGET are treated specially. It does not make # sense to expand them in the global scope since they will never be defined # there under normal circumstances. Therefore they expand to a string that # will later be expanded correctly, when the variable is evaluated again in # the scope of an actual target. # # Even though the "@" variable is not defined at this point, this is not an # error. In all practical cases, this is no problem. This particular test # case is made up and unrealistic. .if $@ != "\$(.TARGET)" . error .endif # Since 2020-09-13, Var_Parse properly reports errors for undefined variables, # but only in lint mode. Before, it had only silently returned var_Error, # hoping for the caller to print an error message. This resulted in the # well-known "Malformed conditional" error message, even though the # conditional was well-formed and the only error was an undefined variable. # expect+2: Malformed conditional (${UNDEF}) # expect+1: Variable "UNDEF" is undefined .if ${UNDEF} . error .endif # Since 2020-09-14, dependency lines may contain undefined variables. # Before, undefined variables were forbidden, but this distinction was not # observable from the outside of the function Var_Parse. ${UNDEF}: ${UNDEF} # In a condition that has a defined(UNDEF) guard, all guarded conditions # may assume that the variable is defined since they will only be evaluated # if the variable is indeed defined. Otherwise they are only parsed, and # for parsing it doesn't make a difference whether the variable is defined # or not. .if defined(UNDEF) && exists(${UNDEF}) . error .endif # Since 2020-10-03, in lint mode the variable modifier must be separated # by colons. See varparse-mod.mk. # expect+2: Missing delimiter ':' after modifier "L" # expect+1: Missing delimiter ':' after modifier "P" .if ${value:LPL} != "value" . error .endif # Between 2020-10-03 and var.c 1.752 from 2020-12-20, in lint mode the # variable modifier had to be separated by colons. This was wrong though # since make always fell back trying to parse the indirect modifier as a # SysV modifier. # expect+1: Unknown modifier "${" .if ${value:${:UL}PL} != "LPL}" # FIXME: "LPL}" is unexpected here. . error ${value:${:UL}PL} .endif # Typically, an indirect modifier is followed by a colon or the closing # brace. This one isn't, therefore make falls back to parsing it as the SysV # modifier ":lue=lid". .if ${value:L:${:Ulue}=${:Ulid}} != "valid" . error .endif # In lint mode, the whole variable text is evaluated to check for unclosed # expressions and unknown operators. During this check, the subexpression # '${:U2}' is not expanded, instead it is copied verbatim into the regular # expression, leading to '.*=.{1,${:U2}}$'. # # Before var.c 1.856 from 2021-03-14, this regular expression was then # compiled even though that was not necessary for checking the syntax at the # level of variable expressions. The unexpanded '$' then resulted in a wrong # error message. # # This only happened in lint mode since in default mode the early check for # unclosed expressions and unknown modifiers is skipped. # # See VarCheckSyntax, ApplyModifier_Regex. # VARMOD_REGEX= ${:UA=111 B=222 C=33:C/.*=.{1,${:U2}}$//g}