This gives a considerable speedup in the processing of .WAIT and .ORDER.
Both .WAIT and .ORDER stop both the commands of the node, and its dependant
nodes being built until the LH nodes are complete.
.WAIT only applies to the dependency line on which it appears, whereas
.ORDER applies globally between the two nodes.
In both cases dependant nodes can be built because other targets need them.
make now processes the target list left to right, scheduling child nodes
as they are needed to make other nodes (instead of attempting to generate
a bottom-up dependency graph at the start). This means that 'make -j1'
will tend to build in the same order as a non-parallel make.
Note that:
all: x y
x: a .WAIT b
y: b .WAIT a
does not generate a dependency loop.
But
x: y
.ORDER y x
does (unless something elswhere causes 'y' to be built).
(Almost all the debug output went there, but some went to stderr.)
Split the parsing of -d (debug flags) out into its own routine.
Allow the output filename to be changed by specifying -dF<file> to create
a log file, or -dF+<file> to append to it. <file> may be stdout or stderr.
Also change so that -d-<flags> acts on <flags> locally but doesn't copy
them to MAKEFLAGS so they aren't inherited by child makes.
I'm not 100% happy with the command line syntax for the above, so they are
currently undocumented.
entries, so allocate at startup.
Use an 'int jobPipe[2]' within the job structure, and create pipes directly
into it. Common up the code that creates all the pipes - making them all
non-block on the read side in the process.
Call Job_CatchChildren() directly from Job_CatchOutput() so that it only
gets called when a child actually exits.
NB: Something causes a 'pregnant pause' if (for example) you call 'nbmake obj'
in src/tools. Introduced between netbsd 3 and 4.
- Send each type of signal to its own handler.
- Only call JobFinish when a process exits, in particular don't 'fake up'
'exitstatus' for jobs being continued, nor call it for suspends.
- When a job is stopped, use an entire variable to remember the fact, so
we know we need to send a SIGCONT. Don't change any other state.
- In order to report '*** [job3] Suspended' before we suspend ourselves we
have to call waitpid() from the signal handler - where we don't want to
process job termination events. Save the exit status and process later.
The code now handles:
- jobs that suspend themselves
- jobs exiting while suspended
- jobs that don't actually suspend at all
Hoewever it still does printfs() from the signal handler, and I haven't yet
stopped it thrashing the signal mask.
De-optimise the getting of a job token so we don't re-order the job
list when there are no tokens.
This might have helped etc/Makefile, but isn't enough.
Instead of adding MAKE_BOOTSTRAP for hosted environments, i.e., when
you want things simple, instead add MAKE_NATIVE to get those hugely
important features like __RCSID().
It's now possible to build make on some hosts with: cc *.c */*.c
In these revisions `::' dependency handling was simplified by not linking
the cohort nodes into the dependency graph. This broke dependency checking
on all but the first instance of a `::' target since all of the cohort nodes
now just form a collection of disconnected dependency graphs.
Fix this by keeping a back-reference in each cohort to its leader (the
first instance of a :: node with the same name) and a count of the number
of cohorts that need to be made before dependent nodes are scheduled.
Classically, we'd need six centurions for cohort, but in this case one
suffices...
the removal of .USE nodes from a node's children list in case a given .USE
nodes appears multiple times on that list, preventing the target from ever
making it on the `to be made' queue.
Since the suffix rule processing code deals itself with removing the
transformation nodes from the parents to which it applies them, arrange
for doing the same for .USE nodes in MakeHandleUse() instead of in
Make_HandleUse(), and still use the OP_MARK stuff to avoid duplication
of commands.
Also, since Make_HandleUse() is not a list callback function and its
return value is no longer used, make its return type void.
normal target (the latter dependency being the reason why it's made),
the .MADE parent should not be put on the `toBeMade' list again (in
Make_Update()) since it was already put on it in Make_ExpandUse().
Doing so would cause the rules for the .MADE target to be executed
(at least) twice, and also mess up the unmade children count of _its_
parent, resulting in spurious graph cylce detection.
To achieve this, make sure the unmade children count of the .MADE target
is set to zero before enqueueing it on the `toBeMade' list in
Make_ExpandUse(). Then, in Make_Update(), check if the parent has the
.MADE attribute before diddling with the queue.
For the same reason the status of a node must not be set to UPTODATE
unconditionally in compat mode, since that will prevent the node from
being built even if it is the source of a normal target. Instead,
check both its state and the type of the parent to decide whether to
continue on behalf of the current parent.
the .DOTLAST primitive by a boolean variable with the same name, this whole
mechanism was broken ... it doesn't save much stat calls but it was wrong.
Thanks to Jason Thorpe for the other shadow-variable fixing patches he
made.
Add -N flag to *really* not execute any commands (useful when using
the -d flags to debug usr/src/Makefile)
Document -N
Update documentation of -n to mention that it still executes commands
for targets marked .MAKE so that the -N/-n distinction is clear.
use -j; all make's in a recursive build cooperate to limit the total
number of jobs, using a token-passing scheme.
The current token passing algorithm is similar to the one implemented
by gmake; there is a single pipe which is inherited through the entire
process hierarchy; tokens are obtained by reading a byte from the
"read end" of the pipe, and are returned by writing them to the "write
end". This exact algorithm is likely to change in the future.
Implementation details:
- Use the new trace facility to allow measurement of the
effectiveness of different token-passing schemes
- Get a token in MakeStartJobs(), return it in Make_Update()
- Eliminate Job_Full() and the jobFull global since they are
redundant with token system.
- Add an "internal" -J option (to pass the token pipe fd's down to
submakes) and a -T option for tracing.
- Change how compatMake is forced so that -j means something when
inherited by submakes.
- When waiting for a token, poll the token-passing pipe as well as
the output pipes of existing jobs.
arguments names on one function being swapped (by a previous author).
Do not do any duplicate suppression when a source list is created. Instead:
* OP_MADE protects against trying to make the source multiple times.
* A new OP_MARK flag is introduced to suppress duplicates while expanding
the .ALLSRC variable and .USE targets.
This turns the O(n^2) insertion into O(n) in most cases.
This is tested with a `make build' and some special test cases.
christos