Error handling: Provide err and ddbl_only, and an internal spec

No callers of these yet.  The spec is honoured in the breach.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Make @nest always be nonempty

No outward functional change, but makes the error handling
etc. slightly simpler.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Do not tolerate unclosed nested scopes at EOF

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Macro assistance part 2 - syntax for $(eval $(call...))

This can be used to call a macro defined with &:macro.  In combination
these provide a more reasonable macro facility than make's.

Specifically:

* Document the new facility.
* Motivate it in the README.
* Provide the new @nest kind for &${  }, which uses $ddbl as
  a counter to count { and }.
* Implement the {}-counting and nesting end in the main parsing
  loop.  (We must now search for things besides $esc.)
* In extract-doctest, provide a slightly adhoc special case for
  understanding ${eval ...} as an expansion text in the README.
  And do not attempt to test &${ } when already $-doubling.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Macro assistance part 1 - macro directive

Define the &:macro directive, which is a dollar-doubled version of
make's own `define'.

We must introuce a new concept of `nesting', for when "the content of
the construct is $-doubled" in the words of the README.  The
implementation is in the main process_input_mk function: it is a
multi-line scope of some kind with an ad-hoc ending condition, and its
own dollar-doubling setting.

We will also want a convenient syntax for $(eval $(call...)), which
we're going to introduce in a moment.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

doctests: Rename `desc' etc. in extract-doctest from `rubric'

This isn't a rubric, it is a description of a particular test entry.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

doctests: Improve entry description filter

This makes the decriptive text more aposite in some situations.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Dollar doubling feature

This makes it easier to write certain rules commands, and macros
involving eval.  Further enhancements will follow, but for now:

* Document the new facility, including the &-escapes to turn it on and
  off, its effect, and the three within-doubling don't-double escapes.

* Implementation.  Specifically:
   - Rename od to oud (`undoubled').
   - Provide od which does double things if $ddbl (a global) is set.
   - Implement the escapes, and add some error checks.
   - Change `od' to `oud' in `&\$', so that that always produces an
     undoubled $.

* Enhance extract-doctest:
   - Set $e->{DD} to say if it's in `while dollar doubling'.
   - Filter out such entries from the `normal' test - they are not
     legal outside dollar-doubling.
   - Add a dollar-doubling test which includes versions of all
     the normal tests as well as the dollar-doubling ones, and
     which dollar-doubles the outputs when appropriate.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/filter/.gitignore: Ignore all .tmp

Rather than trying to list them all.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Introduce oraw

Replace all other calls to `o' with `oraw'.  This ensures we
have no stray calls.

It elminates an anomaly, namely that calls outside process_input_mk
would rely on the $buffering_output variable which is set only there.

No functional change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Introduce od

Replace every call to `o' in process_input_mk (the main filtering
routine) with a call to `od', which is going to be the dollar-doubled
output variant when we add the dollar doubling feature.

For now it just does the same as `o'.

No functional change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

example: Introduce a Final.sd.mk and test it

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/filter: Test Final.sd.mk

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Interface: Read Final.sd.mk

Tests will come in a moment, although we already see the new output
file mentioned in main.mk.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

README: fix TARGETS text for new tidier semantics

In
  Syntax: Refine &TARGETS_things, mostly to avoid # wrinkle
we updated the spec for &TARGETS_things in the list of expansions,
but not in the discussion of per-directory recursive targets.

Fix that.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Interface: Rename Perdir to Suffix; provide Prefix

We want to be able to specify something that appears at the top of
every per-directory segment, principally to provide
for-every-directory variable settings.

A more sophisticated scheme akin to m4 diversions would be possible in
principle, but it would make organising the output files (which must
be processed by autoconf) really awkward - let alone the depenency
management for regenerating the makefiles.

So instead, simply provide a Prefix too.  Additionally, soon we are
going to introduce a Final.sd.mk which is processed right at the end.

I think this new scheme will be sufficient in principle to do anything
that might be needed.  Sadly it will continue to require that
&:include's are done in the right order.

"Suffix" and "Subdir" tab-complete really badly.  In a forthcoming
commit we will rename Subdir.sd.mk to Dir.sd.mk.

Incompatible change.  We must update the tests right away.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Make many dir vars globals

This makes the code a bit more fragile but a lot less repetitious.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Fix an erroneous comment

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Drop comment about input files

This is all documented in README now.
And it's about to become out of date.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/check: Dynamic parallism choice

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/check: Show parallel make rune we run

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests: Run in parallel

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/example: Reset times of (copied) source files too

This eliminates a problem with our test case: if the build takes less
than 1s the files were probably copied more recently than the 1s ago,
resulting in too much rebuilding (and possibly missing
failed-to-rebuild bugs).

We can now change the timestamp for the build tree to something
earlier.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/example: Build in a copy

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/build-common: Copy subdirmk, rather than symlinking

This is more like it will be in someone's project.  Also it prevents
us from accidentally touching anything in the source tree.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/intree: Make copy and break out from tests/example

We introduce make_copy, which makes a copy of tests/example
suitable for testing.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

build tests: Break out build-common

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

regen.mk.in: Clarify documentation about realclean

It's not sensible to `make [the provided] realclean:: into a recursive
target' in that sense since the one we provide is effective, and
present, only in the toplevel.  What we mean is that you can make your
own realclean target and regen.mk's will fit in nicely.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

regen.mk.in: Document CONFIG_STATUS_OUTPUTS

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test/example/check: Reset times for regen tests

In principle, on a filesystem with poor time resolution, the updates
to the input files might have the same timestamp as the output files.
Things would then not be regenerated.

Work around this by resetting all timestamps in the build tree to
1s ago.  This should ensure that exactly the one file whose edit we
are simulating seems, to make, to have been changed.

There is a remaining possible problem: if our last actual change to
the subdirmk source tree is less than 1s ago, this might rebuild too
much.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

test/example/check: Make sections appear in stderr output

This makes the output easier to navigate.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/example/check: Better handling of makefile updates

Rather than checking just for presence of a string in a templated
output file, also have the input file define a target which we try to
run.  This checks that make actually reread the file.

Also add ||false to the grep lines, which makes the stderr output
more obviously contain a failure when the test fails.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/example: Include non-.sd.mk files in SUBDIRMK_MAKEFILES

subdirmk.ac provides @_SUBDIRMK_MAKEFILES@ which is a list of the
*.mk makefiles (corresponding to *.mk.in makefiles) passed to
SUBDIRMK_MAKEFILES.  These need to be included in the
identically-named make variable for regen.mk to rerun config.status
when the .in files are edited.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/example: Correct filename for for-test.mk

&^/ means "this source directory" so &^/lib/ in lib/ is nonsense and
produces lib/lib/.  Also we didn't want the source directory:
anyway: config.status produces outputs in the build tree.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

regen.mk.in: Debugging, controlled by SUBDIRMK_REGEN_NDEBUG

And enable this in test/example/check.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests: Test Perdir.sd.mk

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

README: Do not assert that all directive arguments are &-expanded

Since
  Syntax: only some directives' RHS should be &-expanded
this has not been true for &:changequote, and it is not true in
the general case.  It depends on the directive.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

README: Document NEWQUOTE\$ in changequote

This also adds a doctest.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Syntax: Line joining with &\

make has a weird rule for line joining in recipes, where the \ is
retained.  This is very annoying espcially when writing little scripts
in other languages in recipes.  The GNU make documentation suggests
using a `define' block, in which the \ are still necessary, but
vanish.  But that separates out the recipe from the rule, and also
demands that the recipe text be given a name.

(Note that .ONESHELL offers an alternative but sadly is not available
on a per-rule basis.  This is even less sensible with non-recursive
make, where you might want different directories to have different
settings.)

So, provide &\ which does the other thing, swallowing the newline and
discarding the &\, even in rules.

Add a test case too.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests: Test &TARGETS feature

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Syntax: Refine &TARGETS_things, mostly to avoid # wrinkle

We will add a test case in a moment, since it is quite a noisy diff.

We have to upate clean.sd.mk, which uses this feature.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Syntax: only some directives' RHS should be &-expanded

In principle this is an incompatible change but only for strange input
files and only if they use &:changequote.

We need an ad-hoc test for this.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Syntax: Provide &$VARIABLE

Testing is done by the doctests.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

doctests: Test in sub/dir too

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

doctests: Add missing & from include

This makes no difference in this file but we'll want the same in
sub/dir/ and they should be the same.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

doctests: check: Define i and o vars

This makes bash -x debugging easier.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

doctests: Fail on extraction if we don't understand the README

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/filter/check: Enable pipefail

So we must use bash.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

doctests: Extract and use

Include the doctests from the toplevel Subdir.sd.mk.

In check, extract the doctests, and seddery the expected output into
the combined expected output file.  This involves adding a comment in
Subdir.mk.expected to indicate where the doctests will appear.

In update-expected, contrive to strip the doctest output back out of
the actual output so we can update only the manually-edited part.

Consequential changes to MAKEFILE_TEMPLATES in main.mk.expected
and .gitignore.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

doctests: Fix newline handling in changequote mode section

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

README, doctests: Use NEWQUOTE instead of STUFF

STUFF is rather generic.  Also we want to seddery it in the rubric for
the doctest to avoid it triggering.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

doctests: Provide extract-doctest

This is not called yet.  There are a few things that need fixing
first.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

README: Do not use => inappropriately

Our notation is that => introduces the ouput string.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Syntax: Use &\ for escaping; abolish \&

"\&" causes lots of trouble, including difficulty specifying certain
resulting strings.

Previously I proposed a scheme using &$@ but this is only useful to
provide a way to write &s.

The $-doubling regime which will come later wants a way to produce a
single literal $, so introduce that now using the same syntax.
And &\ feels quite natural to me.

Incompatible change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Syntax: Introduce &. and friends

Incompatible change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Provide $var_prefix_name to process_input_mk

Nothing uses this yet.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Syntax: Introduce &:changequote

Incompatible change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Track output Makefiles properly

This fixes a bug where the intermediate directory Makefile was not
included whe it should be.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/filter: Add .../sub/dir

This will let us test output which depends on the subdirectory,
which is kind of the whole point...

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests: Shuffle, make tests/check run all the tests

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests/filter: New test framework for generate

This test case bypasses autoconf and runs generate directly.
It also gives us somewhere to write examples of each of the
&-expansions without further cluttering test-example.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Handle nodes with no own Subdir.sd.mk

If a directory is not mentioned on the command line, we would assume
it had a Subdir.sd.mk anyway.  Instead, expect it not to have one.

We do still generate a per-subdirectory fragment for such unmentioned
nodes: this serves to wire its children ino the parent, make running
`make' in that directory possible, etc.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

generate: Add missing final newline to main.mk

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

tests: Add some section comments.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

example: Add a comment next to for-test bits

This may help unconfuse people using this as an example.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

README: Fix a mistake in the file reference table

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Merge from secnet

&TARGETS*: make autogenerated targets .PHONY

Suggested-by: Mark Wooding <mdw@distorted.org.uk>
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Minor fix to README

Reported-by: Mark Wooding <mdw@distorted.org.uk>
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Minor fixes to regen.mk.in comment

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: Suppress MAKEFILE_TEMPLATES when running `make clean'

See the README hunk for the reasoning.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: break out subdirmk_target in generated Makefile

We are going to want to reuse this.  Make it a simply-expanded make
variable rather than a perl variable, so that it only gets computed
once.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: stub Makefiles: filter out `all'

Otherwise `make all' generates this:
  Makefile:3: target 'all' given more than once in the same rule
which is harmless but mildly irritating.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: clean.sd.mk: Delete .*~ too (eg .gitignore~)

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: cdeps.sd.mk: Honour &CDEPS_OBJECTS

This makes this facility more flexible.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: Fix bug in regen.mk.in

Apparently, if you just write `;', make knows that the command can't
have done anything.  It therefore concludes tha the target file cannot
have been updated and doesn't bother rereading it.

The result is that you end up using last run's makefiles.

Writing @: fixes this.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: Change stub force target to `run-main.mk'

This produces significantly better error messages:
  [some error from actual makefiles]
  make[1]: Leaving directory '/home/ian/wherever'
  Makefile:6: recipe for target 'run-main.mk' failed

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: Fix bug when target in subdir exists

If you try to make a file that exists, the % rule doesn't seem to
match.  But providing the dependency for all of $(MAKCMDGOALS)
works.

We have to mention `all' explicitly because it might not appear in
$(MAKECMDGOALS).

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Docs: Fix some leftover ; and ,

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: Docs, go back to ^ and ~

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: Go back to ^ and ~

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: Implement new syntax

But I plan to go back to ^ and ~

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Discuss variable settings and inclusion order

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

subdirmk: Actually clear right variable

"subdirmk: main.mk: clear SUBDIRMK_MAKEFILES MAKEFILE_TEMPLATES"
cleared MAKEFILES by mistake (this was a rebase transposition error -
this commit originally came before "subdirmk: Rename MAKEFILES
variable".

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Abolish all absolute pathname syntaxes

These are rare and make things much more confusing.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Improve docs.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Fill in an empty cell with a copy instead

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Transpose columns in path syntax reference

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Change . to require rather than imply @

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Further corresponding new syntax

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: wip new syntax

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Swap rows in syntax table

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: New plan for directory references, further

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: New plan for directory references, new

Not yet implemented.  About to simplify this...

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: New plan for directory references

Not yet implemented.  About to simplify this...

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Summarise directory reference syntaxes, permute

Just permute the columns which seems to be clearer.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

docs: Summarise directory reference syntaxes

This reveals it to be a bit of a mess.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Clarify incremental paragraph

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Clarify race lossage paragraph

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Some hints in docs

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>