freebsd-src/unit-tests/cond-func-empty.mk
Simon J. Gerraty cdde9e894d Import bmake-20220204
Features of interest:

Allow setting target local variables (similar to gmake)

In META_MODE .MAKE.META.CMP_FILTER can be used for filtering commands
before comparion.

contrib/bmake/mk/cc-wrap.mk is an example of using these

See ChangeLog for the gory details.
2022-02-05 12:03:50 -08:00

204 lines
6.6 KiB
Makefile

# $NetBSD: cond-func-empty.mk,v 1.17 2021/12/28 22:13:56 rillig Exp $
#
# Tests for the empty() function in .if conditions, which tests a variable
# expression for emptiness.
#
# Note that the argument in the parentheses is a variable name, not a variable
# expression, optionally followed by variable modifiers.
#
.undef UNDEF
EMPTY= # empty
SPACE= ${:U }
WORD= word
# An undefined variable is empty.
.if !empty(UNDEF)
. error
.endif
# An undefined variable has the empty string as the value, and the :M
# variable modifier does not change that.
#
.if !empty(UNDEF:M*)
. error
.endif
# The :S modifier replaces the empty value with an actual word. The
# expression is now no longer empty, but it is still based on an undefined
# variable (DEF_UNDEF). There are a few variable modifiers that turn an
# undefined expression into a defined expression, among them :U and :D, but
# not :S.
#
# XXX: This is hard to explain to someone who doesn't know these
# implementation details.
#
.if !empty(UNDEF:S,^$,value,W)
. error
.endif
# The :U modifier changes the state of a previously undefined expression from
# DEF_UNDEF to DEF_DEFINED. This marks the expression as "being interesting
# enough to be further processed".
#
.if empty(UNDEF:S,^$,value,W:Ufallback)
. error
.endif
# And now to the surprising part. Applying the following :S modifier to the
# undefined expression makes it non-empty, but the expression is still in
# state DEF_UNDEF. The :U modifier that follows only looks at the state
# DEF_UNDEF to decide whether the variable is defined or not. This kind of
# makes sense since the :U modifier tests the _variable_, not the
# _expression_.
#
# But since the variable was undefined to begin with, the fallback value from
# the :U modifier is used in this expression.
#
.if ${UNDEF:S,^$,value,W:Ufallback} != "fallback"
. error
.endif
# The variable EMPTY is completely empty (0 characters).
.if !empty(EMPTY)
. error
.endif
# The variable SPACE has a single space, which counts as being empty.
.if !empty(SPACE)
. error
.endif
# The variable .newline has a single newline, which counts as being empty.
.if !empty(.newline)
. error
.endif
# The following example constructs an expression with the variable name ""
# and the value " ". This expression counts as empty since the value contains
# only whitespace.
#
# Contrary to the other functions in conditionals, the trailing space is not
# stripped off, as can be seen in the -dv debug log. If the space had been
# stripped, it wouldn't make a difference in this case, but in other cases.
#
.if !empty(:U )
. error
.endif
# Now the variable named " " gets a non-empty value, which demonstrates that
# neither leading nor trailing spaces are trimmed in the argument of the
# function. If the spaces were trimmed, the variable name would be "" and
# that variable is indeed undefined. Since CondParser_FuncCallEmpty calls
# Var_Parse without VARE_UNDEFERR, the value of the undefined variable is
# returned as an empty string.
${:U }= space
.if empty( )
. error
.endif
# The value of the following expression is " word", which is not empty.
.if empty(:U word)
. error
.endif
# The :L modifier creates a variable expression that has the same value as
# its name, which both are "VAR" in this case. The value is therefore not
# empty.
.if empty(VAR:L)
. error
.endif
# The variable WORD has the value "word", which does not count as empty.
.if empty(WORD)
. error
.endif
# The expression ${} for a variable with the empty name always evaluates
# to an empty string (see Var_Parse, varUndefined).
.if !empty()
. error
.endif
# Ensure that variable expressions that appear as part of the argument are
# properly parsed. Typical use cases for this are .for loops, which are
# expanded to exactly these ${:U} expressions.
#
# If everything goes well, the argument expands to "WORD", and that variable
# is defined at the beginning of this file. The surrounding 'W' and 'D'
# ensure that CondParser_FuncCallEmpty keeps track of the parsing position,
# both before and after the call to Var_Parse.
.if empty(W${:UOR}D)
. error
.endif
# There may be spaces at the outside of the parentheses.
# Spaces inside the parentheses are interpreted as part of the variable name.
.if ! empty ( WORD )
. error
.endif
${:U WORD }= variable name with spaces
# Now there is a variable named " WORD ", and it is not empty.
.if empty ( WORD )
. error
.endif
# Parse error: missing closing parenthesis.
.if empty(WORD
. error
.else
. error
.endif
# Since cond.c 1.76 from 2020-06-28 and before var.c 1.226 from 2020-07-02,
# the following example generated a wrong error message "Variable VARNAME is
# recursive".
#
# Since at least 1993, the manual page claimed that irrelevant parts of
# conditions were not evaluated, but that was wrong for a long time. The
# expressions in irrelevant parts of the condition were actually evaluated,
# they just allowed undefined variables to be used in the conditions, and the
# result of evaluating them was not used further. These unnecessary
# evaluations were fixed in several commits, starting with var.c 1.226 from
# 2020-07-02.
#
# In this example, the variable "VARNAME2" is not defined, so evaluation of
# the condition should have stopped at this point, and the rest of the
# condition should have been processed in parse-only mode. The right-hand
# side containing the '!empty' was evaluated though, as it had always been.
#
# When evaluating the !empty condition, the variable name was parsed as
# "VARNAME${:U2}", but without expanding any nested variable expression, in
# this case the ${:U2}. The expression '${:U2}' was replaced with an empty
# string, the resulting variable name was thus "VARNAME". This conceptually
# wrong variable name should have been discarded quickly after parsing it, to
# prevent it from doing any harm.
#
# The variable expression was expanded though, and this was wrong. The
# expansion was done without VARE_WANTRES (called VARF_WANTRES back
# then) though. This had the effect that the ${:U1} from the value of VARNAME
# expanded to an empty string. This in turn created the seemingly recursive
# definition VARNAME=${VARNAME}, and that definition was never meant to be
# expanded.
#
# This was fixed by expanding nested variable expressions in the variable name
# only if the flag VARE_WANTRES is given.
VARNAME= ${VARNAME${:U1}}
.if defined(VARNAME${:U2}) && !empty(VARNAME${:U2})
.endif
# If the word 'empty' is not followed by '(', it is not a function call but an
# ordinary bare word. This bare word is interpreted as 'defined(empty)', and
# since there is no variable named 'empty', the condition evaluates to false.
.if empty
. error
.endif
empty= # defined but empty
.if empty
.else
. error
.endif