freebsd-nq/contrib/bmake/unit-tests/var-op-default.mk
Simon J. Gerraty 06b9b3e0ad Merge bmake-20210110
Quite a lot of churn on style, but lots of
good work refactoring complicated functions
and lots more unit-tests.
Thanks mostly to rillig at NetBSD

Some interesting entries from ChangeLog

o .MAKE.{UID,GID} represent uid and gid running make.

o allow env var MAKE_OBJDIR_CHECK_WRITABLE=no to skip writable
  checks in InitObjdir.  Explicit .OBJDIR target always allows
  read-only directory.

o add more unit tests for META MODE

Merge commit '8e11a9b4250be3c3379c45fa820bff78d99d5946' into main

Change-Id: I464fd4c013067f0915671c1ccc96d2d8090b2b9c
2021-01-13 22:21:37 -08:00

78 lines
2.8 KiB
Makefile

# $NetBSD: var-op-default.mk,v 1.3 2020/12/07 21:35:43 rillig Exp $
#
# Tests for the ?= variable assignment operator, which only assigns
# if the variable is still undefined.
# The variable VAR is not defined yet. Therefore it gets the default value
# from the variable assignment.
VAR?= default value
.if ${VAR} != "default value"
. error
.endif
# At this point, the variable 'VAR' is already defined. The '?=' therefore
# ignores the new variable value, preserving the previous "default value".
VAR?= ignored
.if ${VAR} != "default value"
. error
.endif
# The '?=' operator only checks whether the variable is defined or not.
# An empty variable is defined, therefore the '?=' operator does nothing.
EMPTY= # empty
EMPTY?= ignored
.if ${EMPTY} != ""
. error
.endif
# The .for loop is described in the manual page as if it would operate on
# variables. This is not entirely true. Instead, each occurrence of an
# expression $i or ${i} or ${i:...} is substituted with ${:Uloop-value}.
# This comes very close to the description, the only difference is that
# there is never an actual variable named 'i' involved.
#
# Because there is not really a variable named 'i', the '?=' operator
# performs the variable assignment, resulting in $i == "default".
.for i in loop-value
i?= default
.endfor
.if ${i} != "default"
. error
.endif
# At the point where the '?=' operator checks whether the variable exists,
# it expands the variable name exactly once. Therefore both 'VAR.param'
# and 'VAR.${param}' expand to 'VAR.param', and the second '?=' assignment
# has no effect.
#
# Since 2000.05.11.07.43.42 it has been possible to use nested variable
# expressions in variable names, which made make much more versatile.
# On 2008.03.31.00.12.21, this particular case of the '?=' operator has been
# fixed. Before, the '?=' operator had not expanded the variable name
# 'VAR.${:Uparam}' to see whether the variable already existed. Since that
# variable didn't exist (and variables with '$' in their name are particularly
# fragile), the variable assignment with "not used" was performed, and only
# during that, the variable name was expanded.
VAR.param= already defined
VAR.${:Uparam}?= not used
.if ${VAR.param} != "already defined"
. error
.endif
# Now demonstrate that the variable name is indeed expanded exactly once.
# This is tricky to measure correctly since there are many inconsistencies
# in and around the code that expands variable expressions in the various
# places where variable expressions can occur. If in doubt, enable the
# following debug flags to see what happens:
#.MAKEFLAGS: -dcpv
EXPAND_NAME= EXPAND.$$$$ # The full variable name is EXPAND.$$
PARAM= $$$$
EXPAND.${PARAM}?= value with param
.if ${${EXPAND_NAME}} != "value with param"
. error
.endif
.MAKEFLAGS: -d0
all:
@:;