I made a mistake while reading the Posix spec. In fact, the implicit
prerequisite imposed by the .c.o-rule is evaluated as expected, even if
the .c is not given in the target rule.
This partially reverts 0efd64ffaa.
This drops a lot of code and even though we are using a common function,
the specification of only the needed things keeps it very efficient and
even more efficient in memory and time than the previous solution.
There was a long tinkering process at farbfeld about this, but the sad
truth is that it's the only way to make the Makefile truly portable.
Listing it just as
$(COM:=.o): config.mk $(REQ:=.h)
omits the dependency on the c-file itself, which incurs that strictly
speaking the object file is not depending on the source file, which is
nonsense.
You don't see strictly Posix compliant Makefiles around very often and
most use nasty GNU-extensions everywhere. It is a good idea to go ahead
as a fitting example and show how to write them portably.
- Get rid of camel-casing
- Don't use all-caps for variable names
- use LEN()-macro
- use strncmp() rather than strstr() for prefix-checking
- clean up the tokenizer-loop and don't use copies
- make the loop more readable by separating different breaking
conditions
- stricter error-checking and cleanup
- store the layout directly with bprintf rather than having
a separate buffer
Even if we extend the flag to be an n-flag with a numerical input, we
already have other Unix-tools to take care of us much more flexibly,
e.g. with head(1) you can easily get the first 10 outputs with
slstatus -s | head -n 10,
but also e.g. discard the first one and then get the consecutive 10
outputs.
For the X11-root-window-name, you can limit the runtime with timeout(1)
or a comparable solution.
This reverts commit fa8b939990.
It is not necessary to copy memory or anything. Just keep a pointer to
the active prefix-array and assign the length of the arrays to a
variable.
Make the code more readable by using a switch, be more strict when an
invalid base is passed to it and fix a small oversight in the bottom of
the code where the base 1024 was forgotten to generalized.
These functions take the raw number and a unit and automatically
print it out "scaled down" to a proper SI-prefix, for powers of 2
and 10 respectively.
Apply them to the 2-power cases and keep the 10-power for a later
commit.
First dividing by interval before multiplying with 1000 decreases the
precision by +-(interval - 1) * 1000, as interval arithmetic always
applies the Gauß-function to the result.
This is not necessary and simply reordering the operations mitigates
this.
Within the components, snprintf() was unchecked and had inefficient
calls in some places.
We implement esnprintf() that does all the dirty laundry for us
and use it exclusively now.
Given slstatus is a tool that runs in the background, most likely run
from .xinitrc, it's important to prepend the name of the tool to error
messages so it becomes clear where the error is coming from.
To make this much more consistent, this commit adds warn() and die()
utility functions consistent with other suckless projects and adapts all
calls to fprintf(stderr, *) to the warn() and die() functions, greatly
increasing the readability of the code.
1) Remove setlocale() (locales are harmful and any 'issues' shall
be fixed in different ways that are expected).
2) Disable buffering on stdout with setbuf() rather than flushing
it each time.
3) Make error messages more consistent.
4) Add error checks where applicable.
5) Make code a bit more readable where res is assigned.
6) Use XFlush() rather than XSync() (we don't need to wait for the
XServer to react, which could lead to long hangs on our side).
Posix guarantees that the resulting string is null-terminated, even if
we have an overflow. Instead of doing what has already been done,
properly warn when there has been an error or overflow, so the user can
do something about it.
Be consistent with brackets and add a comment for OpenBSD to make it
clearer what the actual intent of the uncommented line is.
In the Makefile, add a dependency of slstatus.o from config.mk.
Granted, this style is definitely not common, but for the short
utility-functions of this program it's just the right choice. This
provides great flexibility, such that in the long run, it will be
possible to also share code between the OS-implementations.
This also keeps the state-keeping at a minimum and makes it clearer
which functions are implemented on which OS without having to jiggle
around with too many files in the process.
this reverts the commits from 92ab9ef52e up to
d42870d6ca.
After heavy consideration, the component split has more disadvantages
than advantages, especially given there will be utility-functions
sharing quite a lot of code that would then need to be duplicated, as it
does not fit into the util.c due to its speciality.
One big advantage of the component-wise build is readability, and
without doubt, this was achieved here. This point will be addressed
with a different approach that will be visible in the upcoming commits.
One big disadvantage of the component build is the fact that it
introduces state to the build process which is not necessary. Before its
introduction, the only influencing factors where the system-defines
__linux__ and __OpenBSD__. With the components, we are also relying on
the output of uname(1).
Additionally, if the os.mk is not present, make gives the output
$ make
Makefile:5: os.mk: No such file or directory
make: *** No rule to make target 'os.mk'. Stop.
This could easily be fixed by providing some sort of meta-rule for this
file, however, it indicates the problem we have here, and this entire
statefulness will heavily complicate packaging of this tool and makes
the build process much more complex than it actually has to be.
The HDR variable is used for general purpose headers, which are not
given in this project. arg.h and config.h are only needed for
slstatus.c, not every component $(COM:=.c).
This also fixes a problem introduced in
e04a385364, where config.h is pulled into
the tarball.
We remove the hack with the sleep and global "delay"-variable and use a
monotonic clock to derive the desired monotonic properties of the tool.
Inside each function that demands a personal delay we can then just do a
nanosleep() and be done with it.
It's a shame that timespec is so ugly to work with, but there's really
no way to make it more beautiful. However, at this cost though we
finally can set the interval times in milliseconds and not only just
seconds.
We remove setlocale(), because nothing good ever came out of this
function.
Besides that we have some more code refactoring, especially in the
argument loop which saves us a bit of complexity.
d-flag: There's no need for that, use the &-operator or fork+exec in a
non-shell-context. In the latter case you get the PID for free.
v-flag: If you want to find out which version of a package is installed,
consult your package manager. That's his job.
