On 10/15/19 9:55 AM, Daniel P. Berrangé wrote:
On Tue, Oct 15, 2019 at 09:42:45AM -0300, Daniel Henrique Barboza wrote:
I was hoping to quickly re-send the qemu_driver cleanups I've sent some time ago, now using Glib. I started by attempting to change the first VIR_AUTOUNREF() call in qemu_driver.c to g_autoptr(), which happens to be a virStorageSourcePtr type, then I realized that it wasn't that simple. It should be that simple with this commit:
commit 667ff797e8eb8d82f30ab430216a8d2eef6b915a Author: Daniel P. Berrangé <berrange@redhat.com> Date: Fri Oct 4 17:14:10 2019 +0100
src: add support for g_autoptr with virObject instances
we should be able to use g_autoptr for any virObject, without having to lock-step convert to GObject.
What actual problem did you find ?
I failed to notice this commit. Just tried it again and it worked. What happened yesterday was that I attempted to do a simple VIR_AUTOUNREF -> g_autopt replace, faced compile errors and then, since I didn't notice this commit about, I assumed "I guess I need to convert this guy to GObject". In fact, the compile error happened because g_autoptr() does not operate with a 'Ptr' type - something that I learned only during the conversion process. Well, hopefully this patch can serve as a baseline for a future conversion for this object type. Guess I can go back safely to re-send the cleanup patches tha are already pending in the ML hehehe
Following up the instructions found on commit 16121a88a7, I started the conversion. Then 'make check' started to fail because some calls to virObjectRef/virObjectUnref were still remaining in the code, messing up stuff related with mirrorChain in qemu_blockjob.c. Turns out it was easier to burn through all the instances and change them to use GLib. Yes, if you convert from virObject to GObject, you *must* convert all virObjectRef/Unref calls at that time.
This is being sent as RFC because x-I am aware that docs/hacking.html mentions that we shouldn't mix up certain GLib macros with Libvirt ones, thus I am uncertain of whether I have messed up or not. 'make check' works, did a few sanity checks with libvirtd as well. Yes, the need to not mix g_auto* with VIR_AUTO*, is why I did commit 667ff797e8eb8d82f30ab430216a8d2eef6b915a to let you use g_autoptr with virObject, without first converting to GObject.
What if there are other object types in the same function using the VIR macros? For example, inside qemu_driver.c: qemuDomainBlockCopyCommon: VIR_AUTOUNREF(virQEMUDriverConfigPtr) cfg = virQEMUDriverGetConfig(driver); const char *format = NULL; bool mirror_reuse = !!(flags & VIR_DOMAIN_BLOCK_COPY_REUSE_EXT); bool mirror_shallow = !!(flags & VIR_DOMAIN_BLOCK_COPY_SHALLOW); bool existing = mirror_reuse; qemuBlockJobDataPtr job = NULL; VIR_AUTOUNREF(virStorageSourcePtr) mirror = mirrorsrc; bool blockdev = virQEMUCapsGet(priv->qemuCaps, QEMU_CAPS_BLOCKDEV); bool mirror_initialized = false; VIR_AUTOPTR(qemuBlockStorageSourceChainData) data = NULL; VIR_AUTOPTR(qemuBlockStorageSourceChainData) crdata = NULL; Let's say that I change the virStorageSourcePtr up there to g_autoptr(virStorageSource) mirror = mirrorsrc; As long as there are no VIR macros acting in the 'mirror' variable, is it to use g_autoptr there even when everyone else is using VIR_AUTO* macros? Thanks, DHB
Regards, Daniel