require newer glib2 to enable autofree'ing of stack variables exiting scope
v2 cover require newer glib2 to enable autofree'ing of stack variables exiting scope
Both GCC and CLang support a C extension attribute((cleanup)) which allows you to define a function that is invoked when a stack variable exits scope. This typically used to free the memory allocated to it, though you're not restricted to this. For example it could be used to unlock a mutex.
We could use that functionality now, but the syntax is a bit ugly in plain C. Since version 2.44 of GLib, there have been a few macros to make it more friendly to use - g_autofree, g_autoptr and G_DEFINE_AUTOPTR_CLEANUP_FUNC.
https://developer.gnome.org/glib/stable/glib-Miscellaneous-Macros.html
https://blogs.gnome.org/desrt/2015/01/30/g_autoptr/
The key selling point is that it simplifies the cleanup code paths, often eliminating the need to goto cleanup labels. This improves the readability of the code and makes it less likely that you'll leak memory accidentally.
Inspired by seeing it added to glib, and used in systemd, Libvirt finally got around to adopting this in Feb 2019. Overall our experience with it has been favourable/positive, with the code simplification being very nice.
The main caveats with it are
- Only works with GCC or CLang. We don't care as those are the only two compilers we declare support for.
- You must always initialize the variables when declared to ensure predictable behaviour when they leave scope. Chances are most methods with goto jumps for cleanup are doing this already
- You must not directly return the value that's assigned to a auto-cleaned variable. You must steal the pointer in some way. ie
BAD:
g_autofree char *wibble = g_strdup("wibble")
....
return wibble;
GOOD:
g_autofree char *wibble = g_strdup("wibble")
...
return g_steal_pointer(wibble);
g_steal_pointer is an inline function which simply copies the pointer to a new variable, and sets the original variable to NULL, thus avoiding cleanup.
I've illustrated the usage by converting a bunch of the crypto code in QEMU to use auto cleanup.
Changed on v2:
- Actually commit the rest of the changes to patch 3 so that what's posted works :-) Sigh.
Daniel P. Berrangé (3):
glib: bump min required glib library version to 2.48
crypto: define cleanup functions for use with g_autoptr
crypto: use auto cleanup for many stack variables
configure | 2 +- crypto/afsplit.c | 28 ++++---------- crypto/block-luks.c | 74 +++++++++++-------------------------- crypto/block.c | 15 +++----- crypto/hmac-glib.c | 5 --- crypto/pbkdf.c | 5 +-- crypto/secret.c | 38 ++++++++----------- crypto/tlscredsanon.c | 16 +++----- crypto/tlscredspsk.c | 5 +-- crypto/tlscredsx509.c | 16 +++----- include/crypto/block.h | 2 + include/crypto/cipher.h | 2 + include/crypto/hmac.h | 2 + include/crypto/ivgen.h | 2 + include/crypto/tlssession.h | 2 + include/glib-compat.h | 42 +-------------------- 16 files changed, 78 insertions(+), 178 deletions(-)
Signature · 1 line
2.21.0
PATCH v2 1/3 glib: bump min required glib library version to 2.48
Per supported platforms doc[1], the various min glib on relevant distros is:
RHEL-8: 2.56.1
RHEL-7: 2.50.3
Debian (Buster): 2.58.3
Debian (Stretch): 2.50.3
OpenBSD (Ports): 2.58.3
FreeBSD (Ports): 2.56.3
OpenSUSE Leap 15: 2.54.3
SLE12-SP2: 2.48.2
Ubuntu (Xenial): 2.48.0
macOS (Homebrew): 2.56.0
This suggests that a minimum glib of 2.48 is a reasonable target.
Compared to the previous version bump in
commit e7b3af81597db1a6b55f2c15d030d703c6b2c6ac
Author: Daniel P. Berrangé <[email protected]>
Date: Fri May 4 15:34:46 2018 +0100
glib: bump min required glib library version to 2.40
This will result in us dropping support for Debian Jessie and Ubuntu 14.04.
As per the commit message 14.04 was already outside our list of supported build platforms and an exception was only made because one of the build hosts used during merge testing was stuck on 14.04.
Debian Jessie is justified to drop because we only aim to support at most 2 major versions of Debian at any time. This means Buster and Stretch at this time.
The g_strv_contains compat code is dropped as this API is present since 2.44
The g_assert_cmpmem compat code is dropped as this API is present since 2.46
[1] https://qemu.weilnetz.de/doc/qemu-doc.html#Supported-build-platforms
--- configure | 2 +- crypto/hmac-glib.c | 5 ----- include/glib-compat.h | 42 ++---------------------------------------- 3 files changed, 3 insertions(+), 46 deletions(-)
configure +1 -1
@@ -3636,7 +3636,7 @@ fi ########################################## # glib support probe -glib_req_ver=2.40 +glib_req_ver=2.48 glib_modules=gthread-2.0 if test "$modules" = yes; then glib_modules="$glib_modules gmodule-export-2.0"
crypto/hmac-glib.c +0 -5
@@ -21,12 +21,7 @@ static int qcrypto_hmac_alg_map[QCRYPTO_HASH_ALG__MAX] = { [QCRYPTO_HASH_ALG_MD5] = G_CHECKSUM_MD5, [QCRYPTO_HASH_ALG_SHA1] = G_CHECKSUM_SHA1, [QCRYPTO_HASH_ALG_SHA256] = G_CHECKSUM_SHA256, -/* Support for HMAC SHA-512 in GLib 2.42 */ -#if GLIB_CHECK_VERSION (2 ,42 ,0 ) [QCRYPTO_HASH_ALG_SHA512] = G_CHECKSUM_SHA512, -#else - [QCRYPTO_HASH_ALG_SHA512] = -1, -#endif [QCRYPTO_HASH_ALG_SHA224] = -1, [QCRYPTO_HASH_ALG_SHA384] = -1, [QCRYPTO_HASH_ALG_RIPEMD160] = -1,
include/glib-compat.h +2 -40
64 lines not on this page · raw message
Signature · 1 line
2.21.0
PATCH v2 2/3 crypto: define cleanup functions for use with g_autoptr
In reply to Daniel P. BerrangéAllow crypto structs to be used with g_autoptr, avoiding the need to explicitly call XXX_free() functions when variables go out of scope on the stack.
--- include/crypto/block.h | 2 ++ include/crypto/cipher.h | 2 ++ include/crypto/hmac.h | 2 ++ include/crypto/ivgen.h | 2 ++ include/crypto/tlssession.h | 2 ++ 5 files changed, 10 insertions(+)
include/crypto/block.h +2 -0
@@ -268,4 +268,6 @@ uint64_t qcrypto_block_get_sector_size(QCryptoBlock *block);*/ void qcrypto_block_free (QCryptoBlock *block); +G_DEFINE_AUTOPTR_CLEANUP_FUNC(QCryptoBlock, qcrypto_block_free) +#endif /* QCRYPTO_BLOCK_H */
include/crypto/cipher.h +2 -0
@@ -170,6 +170,8 @@ QCryptoCipher *qcrypto_cipher_new(QCryptoCipherAlgorithm alg,*/ void qcrypto_cipher_free (QCryptoCipher *cipher); +G_DEFINE_AUTOPTR_CLEANUP_FUNC (QCryptoCipher, qcrypto_cipher_free) +/** * qcrypto_cipher_encrypt: * @cipher: the cipher object
include/crypto/hmac.h +2 -0
@@ -65,6 +65,8 @@ QCryptoHmac *qcrypto_hmac_new(QCryptoHashAlgorithm alg,*/ void qcrypto_hmac_free (QCryptoHmac *hmac); +G_DEFINE_AUTOPTR_CLEANUP_FUNC (QCryptoHmac, qcrypto_hmac_free) +/** * qcrypto_hmac_bytesv: * @hmac: the hmac object
include/crypto/ivgen.h +2 -0
@@ -203,4 +203,6 @@ QCryptoHashAlgorithm qcrypto_ivgen_get_hash(QCryptoIVGen *ivgen);*/ void qcrypto_ivgen_free (QCryptoIVGen *ivgen); +G_DEFINE_AUTOPTR_CLEANUP_FUNC(QCryptoIVGen, qcrypto_ivgen_free) +#endif /* QCRYPTO_IVGEN_H */
include/crypto/tlssession.h +2 -0
@@ -160,6 +160,8 @@ QCryptoTLSSession *qcrypto_tls_session_new(QCryptoTLSCreds *creds,*/ void qcrypto_tls_session_free (QCryptoTLSSession *sess); +G_DEFINE_AUTOPTR_CLEANUP_FUNC (QCryptoTLSSession, qcrypto_tls_session_free) +/** * qcrypto_tls_session_check_credentials: * @sess: the TLS session object
Signature · 1 line
2.21.0
PATCH v2 3/3 crypto: use auto cleanup for many stack variables
In reply to Daniel P. BerrangéSimplify cleanup paths by using glib's auto cleanup macros for stack variables, allowing several goto jumps / labels to be eliminated.
--- crypto/afsplit.c | 28 +++++----------- crypto/block-luks.c | 74 +++++++++++++------------------------------ crypto/block.c | 15 +++------ crypto/pbkdf.c | 5 +-- crypto/secret.c | 38 ++++++++++------------ crypto/tlscredsanon.c | 16 ++++------ crypto/tlscredspsk.c | 5 ++- crypto/tlscredsx509.c | 16 +++------- 8 files changed, 65 insertions(+), 132 deletions(-)
crypto/afsplit.c +8 -20
@@ -58,7 +58,7 @@ static int qcrypto_afsplit_hash(QCryptoHashAlgorithm hash, }for (i =0 ; i < hashcount; i++) { -uint8_t *out =NULL ; + g_autofree uint8_t *out = NULL;size_t outlen =0 ;uint32_t iv =cpu_to_be32 (i);struct iovec in[] = {
@@ -79,7 +79,6 @@ static int qcrypto_afsplit_hash(QCryptoHashAlgorithm hash,assert (outlen == digestlen);memcpy (block + (i * digestlen), out, (i == (hashcount -1 )) ? finallen : digestlen); -g_free (out); }return 0 ;
@@ -93,13 +92,12 @@ int qcrypto_afsplit_encode(QCryptoHashAlgorithm hash,uint8_t *out, Error **errp) { - uint8_t *block = g_new0(uint8_t, blocklen); + g_autofree uint8_t *block = g_new0(uint8_t, blocklen);size_t i; -int ret =-1 ;for (i =0 ; i < (stripes -1 ); i++) {if (qcrypto_random_bytes (out + (i * blocklen), blocklen, errp) <0 ) { - goto cleanup; +return -1 ; } qcrypto_afsplit_xor(blocklen,
@@ -108,18 +106,14 @@ int qcrypto_afsplit_encode(QCryptoHashAlgorithm hash, block);if (qcrypto_afsplit_hash (hash, blocklen, block, errp) <0 ) { - goto cleanup; +return -1 ; } }qcrypto_afsplit_xor (blocklen, in, block, out + (i * blocklen)); - ret =0 ; - - cleanup: -g_free (block); -return ret; +return 0 ; }
@@ -130,9 +124,8 @@ int qcrypto_afsplit_decode(QCryptoHashAlgorithm hash,uint8_t *out, Error **errp) { - uint8_t *block = g_new0(uint8_t, blocklen); + g_autofree uint8_t *block = g_new0(uint8_t, blocklen);size_t i; -int ret =-1 ;for (i =0 ; i < (stripes -1 ); i++) { qcrypto_afsplit_xor(blocklen,
@@ -141,7 +134,7 @@ int qcrypto_afsplit_decode(QCryptoHashAlgorithm hash, block);if (qcrypto_afsplit_hash (hash, blocklen, block, errp) <0 ) { - goto cleanup; +return -1 ; } }
@@ -149,10 +142,5 @@ int qcrypto_afsplit_decode(QCryptoHashAlgorithm hash, in + (i * blocklen), block, out); - - ret =0 ; - - cleanup: -g_free (block); -return ret; +return 0 ; }
crypto/block-luks.c +22 -52
187 lines not on this page · raw message
crypto/block.c +5 -10
@@ -299,15 +299,13 @@ static int do_qcrypto_block_cipher_encdec(QCryptoCipher *cipher,QCryptoCipherEncDecFunc func, Error **errp) { - uint8_t *iv; + g_autofree uint8_t *iv = niv ? g_new0(uint8_t, niv) : NULL;int ret =-1 ;uint64_t startsector = offset / sectorsize;assert (QEMU_IS_ALIGNED (offset, sectorsize));assert (QEMU_IS_ALIGNED (len, sectorsize)); - iv = niv ?g_new0 (uint8_t, niv) :NULL ; -while (len >0 ) {size_t nbytes;if (niv) {
@@ -320,19 +318,19 @@ static int do_qcrypto_block_cipher_encdec(QCryptoCipher *cipher, }if (ret <0 ) { - goto cleanup; +return -1 ; }if (qcrypto_cipher_setiv (cipher, iv, niv, errp) <0 ) { - goto cleanup; +return -1 ; } } nbytes = len > sectorsize ? sectorsize : len;if (func (cipher, buf, buf, nbytes, errp) <0 ) { - goto cleanup; +return -1 ; } startsector++;
@@ -340,10 +338,7 @@ static int do_qcrypto_block_cipher_encdec(QCryptoCipher *cipher, len -= nbytes; } - ret =0 ; - cleanup: -g_free (iv); -return ret; +return 0 ; }
crypto/pbkdf.c +1 -4
@@ -69,12 +69,10 @@ uint64_t qcrypto_pbkdf2_count_iters(QCryptoHashAlgorithm hash, Error **errp) {uint64_t ret =-1 ; -uint8_t *out; + g_autofree uint8_t *out = g_new(uint8_t, nout);uint64_t iterations = (1 <<15 );unsigned long long delta_ms, start_ms, end_ms; - out =g_new (uint8_t, nout); -while (1 ) {if (qcrypto_pbkdf2_get_thread_cpu (&start_ms, errp) <0 ) { goto cleanup;
@@ -108,6 +106,5 @@ uint64_t qcrypto_pbkdf2_count_iters(QCryptoHashAlgorithm hash, cleanup:memset (out,0 , nout); -g_free (out);return ret; }
crypto/secret.c +16 -22
@@ -72,10 +72,12 @@ static void qcrypto_secret_decrypt(QCryptoSecret *secret,size_t *outputlen, Error **errp) { - uint8_t *key = NULL, *ciphertext = NULL, *iv = NULL; + g_autofree uint8_t *key = NULL; + g_autofree uint8_t *ciphertext = NULL; + g_autofree uint8_t *iv = NULL;size_t keylen, ciphertextlen, ivlen; -QCryptoCipher *aes =NULL ; -uint8_t *plaintext =NULL ; +g_autoptr (QCryptoCipher) aes =NULL ; + g_autofree uint8_t *plaintext = NULL; *output =NULL ; *outputlen =0 ;
@@ -83,27 +85,27 @@ static void qcrypto_secret_decrypt(QCryptoSecret *secret,if (qcrypto_secret_lookup (secret->keyid, &key, &keylen, errp) <0 ) { - goto cleanup; +return ; }if (keylen !=32 ) {error_setg (errp,"Key should be 32 bytes in length" ); - goto cleanup; +return ; }if (!secret->iv) {error_setg (errp,"IV is required to decrypt secret" ); - goto cleanup; +return ; } iv =qbase64_decode (secret->iv,-1 , &ivlen, errp);if (!iv) { - goto cleanup; +return ; }if (ivlen !=16 ) {error_setg (errp,"IV should be 16 bytes in length not %zu" , ivlen); - goto cleanup; +return ; } aes = qcrypto_cipher_new(QCRYPTO_CIPHER_ALG_AES_256,
@@ -111,11 +113,11 @@ static void qcrypto_secret_decrypt(QCryptoSecret *secret, key, keylen, errp);if (!aes) { - goto cleanup; +return ; }if (qcrypto_cipher_setiv (aes, iv, ivlen, errp) <0 ) { - goto cleanup; +return ; }if (secret->format ==QCRYPTO_SECRET_FORMAT_BASE64 ) {
@@ -124,7 +126,7 @@ static void qcrypto_secret_decrypt(QCryptoSecret *secret, &ciphertextlen, errp);if (!ciphertext) { - goto cleanup; +return ; } plaintext =g_new0 (uint8_t, ciphertextlen +1 ); }else {
@@ -136,8 +138,7 @@ static void qcrypto_secret_decrypt(QCryptoSecret *secret, plaintext, ciphertextlen, errp) < 0) { - plaintext =NULL ; - goto cleanup; + return; } if (plaintext[ciphertextlen - 1] > 16 ||
@@ -145,9 +146,7 @@ static void qcrypto_secret_decrypt(QCryptoSecret *secret,error_setg (errp,"Incorrect number of padding bytes (%d) " "found on decrypted data" , (int )plaintext[ciphertextlen -1 ]); -g_free (plaintext); - plaintext =NULL ; - goto cleanup; +return ; }/* Even though plaintext may contain arbitrary NUL
@@ -158,12 +157,7 @@ static void qcrypto_secret_decrypt(QCryptoSecret *secret, *output = plaintext; *outputlen = ciphertextlen; - - cleanup: -g_free (ciphertext); -g_free (iv); -g_free (key); -qcrypto_cipher_free (aes); + plaintext =NULL ; }
crypto/tlscredsanon.c +6 -10
@@ -34,9 +34,8 @@ static intqcrypto_tls_creds_anon_load (QCryptoTLSCredsAnon *creds, Error **errp) { -char *dhparams =NULL ; + g_autofree char *dhparams = NULL;int ret; -int rv =-1 ;trace_qcrypto_tls_creds_anon_load (creds, creds->parent_obj.dir ? creds->parent_obj.dir :"<nodir>" );
@@ -45,20 +44,20 @@ qcrypto_tls_creds_anon_load(QCryptoTLSCredsAnon *creds,if (qcrypto_tls_creds_get_path (&creds->parent_obj,QCRYPTO_TLS_CREDS_DH_PARAMS , false, &dhparams, errp) <0 ) { - goto cleanup; +return -1 ; } ret =gnutls_anon_allocate_server_credentials (&creds->data.server);if (ret <0 ) {error_setg (errp,"Cannot allocate credentials: %s" ,gnutls_strerror (ret)); - goto cleanup; +return -1 ; }if (qcrypto_tls_creds_get_dh_params_file (&creds->parent_obj, dhparams, &creds->parent_obj.dh_params, errp) <0 ) { - goto cleanup; +return -1 ; } gnutls_anon_set_server_dh_params(creds->data.server,
@@ -68,14 +67,11 @@ qcrypto_tls_creds_anon_load(QCryptoTLSCredsAnon *creds,if (ret <0 ) {error_setg (errp,"Cannot allocate credentials: %s" ,gnutls_strerror (ret)); - goto cleanup; +return -1 ; } } - rv =0 ; - cleanup: -g_free (dhparams); -return rv; +return 0 ; }
crypto/tlscredspsk.c +2 -3
@@ -69,7 +69,8 @@ static intqcrypto_tls_creds_psk_load (QCryptoTLSCredsPSK *creds, Error **errp) { -char *pskfile =NULL , *dhparams =NULL ; + g_autofree char *pskfile = NULL; + g_autofree char *dhparams = NULL;const char *username;int ret;int rv =-1 ;
@@ -139,8 +140,6 @@ qcrypto_tls_creds_psk_load(QCryptoTLSCredsPSK *creds, rv =0 ; cleanup:g_free (key.data); -g_free (pskfile); -g_free (dhparams);return rv; }
crypto/tlscredsx509.c +5 -11
@@ -378,7 +378,7 @@ qcrypto_tls_creds_load_cert(QCryptoTLSCredsX509 *creds, {gnutls_datum_t data;gnutls_x509_crt_t cert =NULL ; -char *buf =NULL ; + g_autofree char *buf = NULL;gsize buflen;GError *gerr;int ret =-1 ;
@@ -420,7 +420,6 @@ qcrypto_tls_creds_load_cert(QCryptoTLSCredsX509 *creds,gnutls_x509_crt_deinit (cert); cert =NULL ; } -g_free (buf);return cert; }
@@ -434,9 +433,8 @@ qcrypto_tls_creds_load_ca_cert_list(QCryptoTLSCredsX509 *creds, Error **errp) {gnutls_datum_t data; -char *buf =NULL ; + g_autofree char *buf = NULL;gsize buflen; -int ret =-1 ;GError *gerr =NULL ; *ncerts =0 ;
@@ -446,7 +444,7 @@ qcrypto_tls_creds_load_ca_cert_list(QCryptoTLSCredsX509 *creds,error_setg (errp,"Cannot load CA cert list %s: %s" , certFile, gerr->message);g_error_free (gerr); - goto cleanup; +return -1 ; } data.data = (unsigned char *)buf;
@@ -457,15 +455,11 @@ qcrypto_tls_creds_load_ca_cert_list(QCryptoTLSCredsX509 *creds,error_setg (errp,"Unable to import CA certificate list %s" , certFile); - goto cleanup; +return -1 ; } *ncerts = certMax; - ret =0 ; - - cleanup: -g_free (buf); -return ret; +return 0 ; }
Signature · 1 line
2.21.0
Re: [Qemu-devel] [PATCH v2 3/3] crypto: use auto cleanup for many stack variables
Quoted 4 lines · Daniel P. Berrangé
Simplify cleanup paths by using glib's auto cleanup macros for stack variables, allowing several goto jumps / labels to be eliminated.
Signed-off-by Daniel P. Berrangé <[email protected]>
Quoted 10 lines
--- crypto/afsplit.c | 28 +++++----------- crypto/block-luks.c | 74 +++++++++++++------------------------------ crypto/block.c | 15 +++------ crypto/pbkdf.c | 5 +-- crypto/secret.c | 38 ++++++++++------------ crypto/tlscredsanon.c | 16 ++++------ crypto/tlscredspsk.c | 5 ++- crypto/tlscredsx509.c | 16 +++------- 8 files changed, 65 insertions(+), 132 deletions(-)
Quoted 59 lines · Daniel P. Berrangé
Both GCC and CLang support a C extension attribute((cleanup)) which allows you to define a function that is invoked when a stack variable exits scope. This typically used to free the memory allocated to it, though you're not restricted to this. For example it could be used to unlock a mutex.
We could use that functionality now, but the syntax is a bit ugly in plain C. Since version 2.44 of GLib, there have been a few macros to make it more friendly to use -
g_autofree,g_autoptrandG_DEFINE_AUTOPTR_CLEANUP_FUNC.https://developer.gnome.org/glib/stable/glib-Miscellaneous-Macros.html
https://blogs.gnome.org/desrt/2015/01/30/g_autoptr/
The key selling point is that it simplifies the cleanup code paths, often eliminating the need to goto cleanup labels. This improves the readability of the code and makes it less likely that you'll leak memory accidentally.
Inspired by seeing it added to glib, and used in systemd, Libvirt finally got around to adopting this in Feb 2019. Overall our experience with it has been favourable/positive, with the code simplification being very nice.
The main caveats with it are
- Only works with GCC or CLang. We don't care as those are the only two compilers we declare support for.
- You must always initialize the variables when declared to ensure predictable behaviour when they leave scope. Chances are most methods with goto jumps for cleanup are doing this already
- You must not directly return the value that's assigned to a auto-cleaned variable. You must steal the pointer in some way. ie
BAD:
g_autofreechar *wibble =g_strdup("wibble")
....
return wibble;GOOD:
g_autofreechar *wibble =g_strdup("wibble")
...
returng_steal_pointer(wibble);
g_steal_pointeris an inline function which simply copies the pointer to a new variable, and sets the original variable to NULL, thus avoiding cleanup.I've illustrated the usage by converting a bunch of the crypto code in QEMU to use auto cleanup.
Changed on v2:
Quoted 7 lines
- Actually commit the rest of the changes to patch 3 so that what's posted works :-) Sigh.
Daniel P. Berrangé (3):
glib: bump min required glib library version to 2.48
crypto: define cleanup functions for use withg_autoptr
crypto: use auto cleanup for many stack variables
Series:
Quoted 23 lines
configure | 2 +- crypto/afsplit.c | 28 ++++---------- crypto/block-luks.c | 74 +++++++++++-------------------------- crypto/block.c | 15 +++----- crypto/hmac-glib.c | 5 --- crypto/pbkdf.c | 5 +-- crypto/secret.c | 38 ++++++++----------- crypto/tlscredsanon.c | 16 +++----- crypto/tlscredspsk.c | 5 +-- crypto/tlscredsx509.c | 16 +++----- include/crypto/block.h | 2 + include/crypto/cipher.h | 2 + include/crypto/hmac.h | 2 + include/crypto/ivgen.h | 2 + include/crypto/tlssession.h | 2 + include/glib-compat.h | 42 +-------------------- 16 files changed, 78 insertions(+), 178 deletions(-)--
2.21.0
Signature · 1 line
Marc-André Lureau
Quoted 5 lines · Daniel P. Berrangé
Both GCC and CLang support a C extension attribute((cleanup)) which allows you to define a function that is invoked when a stack variable exits scope. This typically used to free the memory allocated to it, though you're not restricted to this. For example it could be used to unlock a mutex.
<snip>
Quoted 9 lines
GOOD:
g_autofreechar *wibble =g_strdup("wibble")
...
returng_steal_pointer(wibble);
g_steal_pointeris an inline function which simply copies the pointer to a new variable, and sets the original variable to NULL, thus avoiding cleanup.
Surely this is a particular use case where you wouldn't use g_autofree to declare the variable as you intending to return it to the outer scope?
--
Alex Bennée
Quoted 15 lines · Alex Bennée
Quoted 5 lines · Daniel P. Berrangé
Both GCC and CLang support a C extension attribute((cleanup)) which allows you to define a function that is invoked when a stack variable exits scope. This typically used to free the memory allocated to it, though you're not restricted to this. For example it could be used to unlock a mutex.
<snip>
Quoted 6 lines
GOOD: g_autofree char *wibble = g_strdup("wibble") ... return g_steal_pointer(wibble);
Quoted 6 lines
Quoted 3 lines
g_steal_pointeris an inline function which simply copies the pointer to a new variable, and sets the original variable to NULL, thus avoiding cleanup.Surely this is a particular use case where you wouldn't use
g_autofreeto declare the variable as you intending to return it to the outer scope?
Actually, it's still quite useful if you have intermediate returns:
g_autofree char *wibble = NULL;
if (something)
return NULL;
wibble = g_strdup("wibble");
if (something_else)
return NULL;
return g_steal_pointer(wibble);
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Quoted 15 lines · Alex Bennée
Quoted 5 lines · Daniel P. Berrangé
Both GCC and CLang support a C extension attribute((cleanup)) which allows you to define a function that is invoked when a stack variable exits scope. This typically used to free the memory allocated to it, though you're not restricted to this. For example it could be used to unlock a mutex.
<snip>
Quoted 6 lines
GOOD: g_autofree char *wibble = g_strdup("wibble") ... return g_steal_pointer(wibble);
Quoted 6 lines
Quoted 3 lines
g_steal_pointeris an inline function which simply copies the pointer to a new variable, and sets the original variable to NULL, thus avoiding cleanup.Surely this is a particular use case where you wouldn't use
g_autofreeto declare the variable as you intending to return it to the outer scope?
I think it depends on the situation. Obviously real code will have something in the "..." part I snipped.
You have 20 code paths that can result in returning with an error, where you want to have all variables freed, and only 1 code path for success Then it makes sense to use g_autofree + g_steal_pointer to eliminate many goto jumps.
If you have only 1 error path and 1 success path, then a traditional g_free() call is may well be sufficient.
IOW, as with many coding "rules", there's scope to use personal judgement as to when it is right to ignore it vs folow it.
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Quoted 26 lines · Daniel P. Berrangé
Quoted 15 lines · Alex Bennée
Quoted 5 lines · Daniel P. Berrangé
Both GCC and CLang support a C extension attribute((cleanup)) which allows you to define a function that is invoked when a stack variable exits scope. This typically used to free the memory allocated to it, though you're not restricted to this. For example it could be used to unlock a mutex.
<snip>
Quoted 6 lines
GOOD: g_autofree char *wibble = g_strdup("wibble") ... return g_steal_pointer(wibble);Quoted 6 lines
Quoted 3 lines
g_steal_pointeris an inline function which simply copies the pointer to a new variable, and sets the original variable to NULL, thus avoiding cleanup.Surely this is a particular use case where you wouldn't use
g_autofreeto declare the variable as you intending to return it to the outer scope?I think it depends on the situation. Obviously real code will have something in the "..." part I snipped.
Quoted 7 lines
You have 20 code paths that can result in returning with an error, where you want to have all variables freed, and only 1 code path for success Then it makes sense to use
g_autofree+g_steal_pointerto eliminate many goto jumps.If you have only 1 error path and 1 success path, then a traditional
g_free()call is may well be sufficient.
I suspect this would be worth a write up in HACKING or CODING_STYLE with the next iteration? (which reminds me we should really merge and .rst up those documents)
Quoted 6 lines
IOW, as with many coding "rules", there's scope to use personal judgement as to when it is right to ignore it vs folow it.
Regards,
Daniel
--
Alex Bennée