Bug #71219 [Csd->ReO]: php's configure script incorrectly checks for ttyname_r availability

Edit report at https://bugs.php.net/bug.php?id=71219&edit=1 ID: 71219 Updated by: nikic@php.net Reported by: atoth at atoth dot sote dot hu Summary: php's configure script incorrectly checks for ttyname_r availability -Status: Closed +Status: Re-Opened Type: Bug Package: *Compile Issues Operating System: Linux (Gentoo Hardened) PHP Version: 5.6.16 Block user comment: N Private report: N New Comment: Sorry, I misunderstood the discussion and applied the incorrect fix. Reverted now. However the other proposed fix, using fd 1 instead of 0, does not work for me either. Might it be sufficient to just check for the availability of the ttyname_r symbol? Previous Comments: ------------------------------------------------------------------------ [2016-07-30 16:36:53] nikic@php.net Automatic comment on behalf of nikic Revision: http://git.php.net/?p=php-src.git;a=commit;h=c9d3ff0c6e86398b9471428ab49c6a6fa47ae977 Log: Revert &quot;Fixed bug #71219&quot; ------------------------------------------------------------------------ [2016-07-30 16:25:45] nikic@php.net Automatic comment on behalf of nikic Revision: http://git.php.net/?p=php-src.git;a=commit;h=14d674442ef86ad4e862228a1ff5ecd322ae7759 Log: Fixed bug #71219 ------------------------------------------------------------------------ [2015-12-26 16:46:36] atoth at atoth dot sote dot hu Your logic is right, however the test still returns false. "checking for working ttyname_r() implementation... no, posix_ttyname() will be thread-unsafe" configure.log show this: "configure:79083: checking for working ttyname_r() implementation configure:79104: x86_64-pc-linux-gnu-gcc -o conftest -I/usr/include -O2 -march=native -pipe -pthread -D_REENTRANT -L/usr/lib64 -Wl,-O1 -Wl,--as-needed conftest.c -liodbc -lmcrypt -lltdl -lonig -lstdc++ -lcrypto -lssl -lcrypto -lcrypt -lpam -lgmp -lt1 -lX11 -lXpm -lpng -lz -ljpeg -lvpx -lcrypto -lssl -lcrypto -lenchant -ldb-5.1 -lgdbm -lcurl -lbz2 -lz -lpcre -lcrypto -lssl -lcrypto -lrt -lm -ldl -lnsl -lxml2 -lz -lm -ldl -lcurl -lnghttp2 -lidn -lrtmp -lz -lgmp -lgnutls -lhogweed -lnettle -lssh2 -lssh2 -lssl3 -lsmime3 -lnss3 -lnssutil3 -lplds4 -lplc4 -lnspr4 -lz -lxml2 -lz -lm -ldl -lfreetype -licui18n -licuuc -licudata -licuio -liodbc -liodbcinst -ldl -lodbc >&5 configure:79104: $? = 0 configure:79104: ./conftest configure:79104: $? = 1 configure: program exited with status 1 configure: failed program was: | /* confdefs.h */ | #define PACKAGE_NAME "" | #define PACKAGE_TARNAME "" | #define PACKAGE_VERSION "" -- | #define HAVE_GETPWUID_R 1 | #define HAVE_GETGRGID_R 1 | /* end confdefs.h. */ | | #include <unistd.h> | | int main(int argc, char *argv[]) | { | char buf[64]; | | return ttyname_r(0, buf, 64) ? 1 : 0; | } | configure:79114: result: no, posix_ttyname() will be thread-unsafe" I modified the check to output the return value, which was turned out to be: 25. Errno #25 means: ENOTTY. I modified the test program like this: " #include <unistd.h> #include <stdio.h> #include <errno.h> #include <string.h> int main(int argc, char *argv[]) { char buf[64]; int retval; retval = ttyname_r(0, buf, 64); printf("\nretval 0: %s, fd path 0: %s.\n", strerror(retval), buf); retval = ttyname_r(1, buf, 64); printf("retval 1: %s, fd path 1: %s.\n", strerror(retval), buf); retval = ttyname_r(2, buf, 64); printf("retval 2: %s, fd path 2: %s.\n", strerror(retval), buf); return ttyname_r(0, buf, 64) ? 1 : 0; } " This is the output I got: checking for working ttyname_r() implementation... retval 0: Inappropriate ioctl for device, fd path 0: H1gB?. retval 1: Success, fd path 1: /dev/pts/2. retval 2: Inappropriate ioctl for device, fd path 2: /dev/pts/2. So my logic was wrong, but checking for fd 0 is inappropriate. I guess ti would be better to check for fd 1, instead! What's your opinion? ------------------------------------------------------------------------ [2015-12-26 05:11:46] laruence@php.net I don't understant the problem here, if ttyname_r() return 0; it mean success, then: return ttyname_r(0, buf, 64) ? 1 : 0; return 0, action-if-true is evaluated. seems correct to me. ------------------------------------------------------------------------ [2015-12-25 18:34:19] atoth at atoth dot sote dot hu Description: ------------ While PHP's configure script checks for ttyname_r it uses this code snippet: return ttyname_r(0, buf, 64) ? 1 : 0; Although ttyname returns a pointer to the fd's null-terminated pathname or NULL on error - ttyname_r behaves differently: it stores the pathname in a buffer and returns 0 on success or an error number. http://linux.die.net/man/3/ttyname_r Therefore the above cited check will report a failure upon success. Configure emits a messages about posix_ttyname being thread unsafe, despite ttyname_r is available. I suggest to change the test logic by taking into account the return values. I attach a trivial example patch below. Please note, that this minor issue affects both 5.6* and the new 7.0* branches. I could not select both branches for the report... ------------------------------------------------------------------------ -- Edit this bug report at https://bugs.php.net/bug.php?id=71219&edit=1