casaroli opened a new pull request, #21104:
URL: https://github.com/apache/nuttx/pull/21104
## Summary
Two header changes with the same reason: programs expect these names from
the same headers as with glibc, musl and the BSDs. One commit each.
### systypes-timeval
glibc, musl and the BSDs make `struct timeval` visible through
`<sys/types.h>` (glibc and musl through `<sys/select.h>` in their default
mode), so `<stdlib.h>` also gives it. Programs depend on that: CPython, for
example, declares functions with `struct timeval *` in its internal header
`pycore_time.h` without `<sys/time.h>`. On NuttX the type was then declared
inside the parameter list ("will not be visible outside of this definition"),
and the function definitions conflicted with the declarations.
This PR moves the definition from `<sys/time.h>` to `<sys/types.h>`, which
`<sys/time.h>` includes (and `<sys/select.h>` through it). The layout (`time_t
tv_sec`, `long tv_usec`) does not change.
### unistd-select
Programs written for other systems use `fd_set`, `FD_SETSIZE`, the
`FD_xxx()` macros and `select()` without `<sys/select.h>`: glibc and the BSDs
give them through `<sys/types.h>`, macOS through `<unistd.h>`. The earlier
standard (SUSv2) also lists only `<sys/time.h>`, `<sys/types.h>` and
`<unistd.h>` for `select()`, and commit 7b898450356 made `<sys/time.h>` include
`<sys/select.h>` for this reason. CPython's `selectmodule.c`, for example,
includes only `<unistd.h>` and `<sys/types.h>`, and does not compile ("unknown
type name 'fd_set'", implicit `select()`).
This PR includes `<sys/select.h>` in `<unistd.h>`, as macOS does.
`<sys/types.h>` cannot include it: `<signal.h>` includes `<time.h>`, which
includes `<sys/types.h>` before `sigset_t` and `struct timespec` are defined,
and `<sys/select.h>` needs both. Nothing that `<sys/select.h>` includes
includes `<unistd.h>`, so there is no include cycle.
## Impact
- systypes-timeval: `struct timeval` is complete after `<sys/types.h>` or
`<stdlib.h>`, as on other systems. Code that includes `<sys/time.h>` or
`<sys/select.h>` sees the same type as before. A file that includes
`<sys/types.h>` and defines its own `struct timeval` does not compile any more;
there is no such file in nuttx or nuttx-apps (the only definition is in
`<sys/time.h>`), and the downloaded ESP HAL has none. The definition is inside
the `__ASSEMBLY__` guard of `<sys/types.h>`, which three `.S` files include. No
ABI change. The same in flat, protected and kernel builds; nothing per
architecture. No documentation says where `struct timeval` is defined.
- unistd-select: After `<unistd.h>`, `fd_set`, `FD_SETSIZE`,
`FD_ZERO()`/`FD_SET()`/`FD_CLR()`/`FD_ISSET()`, `select()` and `pselect()` are
visible, and with them the contents of `<signal.h>`, `<time.h>`, `<string.h>`
and `<sys/time.h>`, which `<sys/select.h>` includes. A file that includes
`<unistd.h>` and defines its own name that one of these headers also defines
does not compile any more; there is no such file in nuttx or nuttx-apps (a
search for the functions and the lower-case macros of these headers). No ABI
change. The same in flat, protected and kernel builds; nothing per
architecture. No documentation describes what `<unistd.h>` includes.
## Testing
The branches were first tested one by one (below). Then all commits of this
PR together, on master b79dc18e015: systimeval.c, unistdselect.c (CONFIG_PIPES)
on `qemu-armv8a:knsh` (QEMU with HVF) all PASS, and ostest exits with status 0
on `qemu-armv8a:knsh` and `qemu-armv8a:nsh`.
### systypes-timeval
Test program `systimeval.c` (`types`: a function sets and reads the members
with only `<sys/types.h>` and `<stdlib.h>` included; `same`: `gettimeofday()`
and `select()`, from headers included later, take the same type), compiled
against the export of qemu-armv8a:knsh (kernel build) and run under QEMU (HVF):
| | master aeab6fbf989 | this PR |
|---|---|---|
| `systimeval` | does not compile (invalid use of undefined type 'struct
timeval') | types PASS, same PASS |
ostest exits with status 0 on qemu-armv8a:knsh (kernel) and qemu-armv8a:nsh
(flat).
Header check: each public header of the knsh export (148 headers in
`include/`, `sys/`, `netinet/`, `arpa/`, `net/`) compiled alone, after
`<unistd.h>`, before `<unistd.h>` and after `<sys/types.h>`: the same result as
on master (all compile, except `sys/syscall_lookup.h`, which is not a
standalone header).
Builds, each without new warnings: stm32f4discovery:nsh (flat, Cortex-M4),
lm3s6965-ek:qemu-protected (protected, Cortex-M3), rv-virt:knsh64 (kernel,
RISC-V; its `.S` files include `<sys/types.h>`), rv-virt:citest64 (flat,
RISC-V, with the cmocka and libc test apps) and sim:nsh (macOS host, whose own
`struct timeval` is in the host headers).
### unistd-select
Test program `unistdselect.c` (`types`: `fd_set`, `FD_SETSIZE`, `FD_ZERO()`,
`FD_SET()`, `FD_ISSET()` with only `<unistd.h>` included; `pipe`: `select()` on
a pipe is not readable, then readable after a `write()`), compiled against the
export of qemu-armv8a:knsh (kernel build, with `CONFIG_PIPES`) and run under
QEMU (HVF):
| | master aeab6fbf989 | this PR |
|---|---|---|
| `unistdselect` | does not compile (unknown type name 'fd_set', implicit
`select()`) | types PASS, pipe PASS |
ostest exits with status 0 on qemu-armv8a:knsh (kernel) and qemu-armv8a:nsh
(flat).
Header check: each public header of the knsh export (`include/*.h`, `sys/`,
`netinet/`, `arpa/`, `net/`; 148 headers) compiled alone, after `<unistd.h>`,
before `<unistd.h>` and after `<sys/types.h>`: the same result as on master
(all compile, except `sys/syscall_lookup.h`, which is not a standalone header).
`gcc -H` shows `<unistd.h>` once.
Builds, each without new warnings: stm32f4discovery:nsh (flat, Cortex-M4),
lm3s6965-ek:qemu-protected (protected, Cortex-M3), rv-virt:knsh64 (kernel,
RISC-V), rv-virt:citest64 (flat, RISC-V, with the cmocka and libc test apps)
and sim:nsh (macOS host).
`tools/checkpatch.sh -c -u -m -g` passes.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
