This fixes problem of creation the garbage of zombie processes on POSIX
systems. This also makes behaviour of `popen2` identical in Windows,
namely, all resources are freed only when waiting, giving at the same
time an ability to observe the exit code afterwards.
* [ re #2742 ] Count no. processors online rather than configured
Seems there might be some oddities with what is reported when, e.g.
reporting the maximum number of processors supported by the currently
installed motherboard, regardless of which processor is socketed.
* [ #2754 ] Update CHANGELOG
* [ fix ] Fix returned status of the `system` function
* [ re #238 ] Fix program error condition of `git diff` call in `Golden`
According to documentation, not only negative exit code means error
Previously, the RefC files were located in IDRIS2_PREFIX. This is
decoupled to allow users to change the prefix (for ad-hoc library
install locations, for example).
* Add function that checks whether a file is a terminal device.
* support isTTY function for NodeJS backend.
* don't accidentally interpret 'false' string as truthy number
* less code duplication.
* add `nextDirEntry` which returns `Maybe String`, so `Nothing` on
the end of directory unlike `dirEntry` which returns unspecified error
on the end of directory
* `dirEntry` is deprecated now, but not removed because compiler depends on it
* native implementation of `dirEntry` is patched to explicitly reset `errno`
before the `readdir` call: without it end of directory and error were
indistinguishable
* test added
Operating system counter stores signals as flag set without counter.
So sending two signals to a process may result to one or two signal
handler invocation. Queueing signals inside Idris could give users
false sense of signals being are queue, while they are not.
In particular, test for signal could not work reliably for that
reason.
Also, practically we usually don't need have more than once signal
event.
This is follow-up to #1660. CC @mattpolzin
```
IDRIS2_VERIFY(cond, message_format, ...)
```
When condition is false, crash.
Used in native functions where correct error handling is hard or
not impossible.
For example, `malloc` rarely fails, but if it fails, better crash
with clear error message than spend time debugging null pointer
dereference.
* add `strerror` function
* move `getErrno` to `System.Errno`
* use `strerror` in `Show FileError`
* on node there's no access to `strerror`, so `strerror` just converts the number to string
Pragma once is supported by all compilers for the last ten years.
Better use it instead of include guards (which use different styles
in different files).
Support for simple signal handling was added in
a0a417240e. This commit also adds the
`_simple_handler` function. It seems to me that this function is
intended as a helper function which should only be visible in
`idris_signal.c`, it is not used outside this file. For this purpose it
is probably also marked as inline. However, the inline keyword does not
require the compiler to actually inline the function. As such, the
`_simple_handler` symbol may still be exported if the compiler doesn't
inline the function.
On my system this seems to be the case and causes the following error
during compilation of idris2:
Exception: (while loading libidris2_support.so) Error relocating Idris2-0.4.0/build/exec/idris2_app/libidris2_support.so: _simple_handler: symbol not found
By marking the `_simple_handler` function as `static inline` it is
ensured that the symbol is not exported, thereby preventing the
relocation error.
To be able to eventually refactor/extend `system` function: to be
able to specify a directory, environment variables, specify arguments
as array etc. Ideally it should be something like Rust
[`std::process::Command`](https://doc.rust-lang.org/std/process/struct.Command.html).
This change adds logic to set up sockaddr correctly for connect and
bind, handles the AF_UNIX case for getSockAddr and expands the existing
test to cover unix sockets.
Change the support code for directory creation so that it sets all
permission bits (ugo=rwx). The process's currently active umask will be
subtracted from these permissions, which leads to the result that (for a
standard umask of "022") directories are created with a mode of "0755".
So by default, directories created with `prim__createDir` are now also
group-executable and other-executable by default.
Previous behaviour was to create the directories readable for group and
other, but not executable (mode "744"), and thus inaccessible to anyone
except the owner.
The Network.Socket.Data code previously used hardcoded constants manually read
from auto-generated C source code, however these constants are specific to
Linux. The original code looked like this:
export
ToCode SocketFamily where
-- Don't know how to read a constant value from C code in idris2...
-- gotta to hardcode those for now
toCode AF_UNSPEC = 0 -- unsafePerformIO (cMacro "#AF_UNSPEC" Int)
toCode AF_UNIX = 1
toCode AF_INET = 2
toCode AF_INET6 = 10
The AF_INET6 constant is correct on my Debian 10 laptop:
molly on flywheel ~> grep -rE '^#define AF_INET6' /usr/include
/usr/include/x86_64-linux-gnu/bits/socket.h:#define AF_INET6 PF_INET6
molly on flywheel ~> grep -rE '^#define PF_INET6' /usr/include
/usr/include/x86_64-linux-gnu/bits/socket.h:#define PF_INET6 10 /* IP version 6. */
However, this is not the case on an OpenBSD machine:
spanner# grep -rE '^#define[[:space:]]+AF_INET6' /usr/include
/usr/include/sys/socket.h:#define AF_INET6 24 /* IPv6 */
This commit adds accessor functions to the C runtime support library for
retrieving the values of these macros as they appear in the system libc header
files, which can then be invoked using the normal C FFI machinery.
