On 07/14/2014 03:09 AM, Michal Privoznik wrote:
On 11.07.2014 18:59, Eric Blake wrote:
On 07/11/2014 03:32 AM, Michal Privoznik wrote:
There's this trend in libvirt of using enum types wherever possible. Now that I'm at virSecurityLabelDef let's rework @type item of the structure so we don't have to typecast it elsewhere.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> --- src/conf/domain_conf.c | 6 ++++-- src/security/security_dac.c | 2 +- src/util/virseclabel.h | 2 +- 3 files changed, 6 insertions(+), 4 deletions(-)
I'm not quite as sure about this one. This solves the issue of how to detect errors when parsing strings to enum, but required the use of an intermediate variable which in turn made the patch a net gain in lines of code. If someone forgets to use the intermediate variable for parsing, this backfires. On the other hand, parsing string to enum should be done in just one location, and that's the location touched by this patch. I'm 50-50 on whether to take this, so I'd like someone else to chime in with an opinion.
I hear you. This patch is just a refactor. It does not add anything useful nor solve any issue. It's okay if dropped. But the more I think about our vir*TypeFromString() the more I feel we should do something about it. How about making it follow our typical function return pattern:
int func() {
virMyFavourite x; const char *string;
if (virMyFavouriteTypeFromString(string, &x) < 0) { virReportError("unknown value: %s", string); goto error; }
Yeah, making the API failsafe by separating error return from value setting is what we've done elsewhere (such as virStrToLong_*). It's a bigger, more invasive change, but may be worth it. And if we do it, we can then start sticking enum values inside non-public structs.
That is, we need this diff:
diff --git a/src/util/virutil.c b/src/util/virutil.c index 95d1ff9..40075e9 100644 --- a/src/util/virutil.c +++ b/src/util/virutil.c @@ -407,15 +407,20 @@ virParseVersionString(const char *str, unsigned long *version,
int virEnumFromString(const char *const*types, unsigned int ntypes, - const char *type) + const char *type, + int *val)
} \ - int name ## TypeFromString(const char *type) { \ + int name ## TypeFromString(const char *type, name *val) { \ return virEnumFromString(name ## TypeList, \ ARRAY_CARDINALITY(name ## TypeList), \ - type); \ + type, (int *) val); \ }
# define VIR_ENUM_DECL(name) \
And then a tons of follow up patches. Or even make virEnumString() report the error (that could save a lot of virReportError() calls).
Or even make the error reporting an optional bool parameter. It's also a huge patch; I wonder if we can come up with an alternative name (VIR_ENUM_DECL2 expands to name ## FromString, leaving VIR_ENUM_DECL expanding to name ## TypeFromString - also nice that it's a shorter name since it's usage will be longer), where we can then focus on just a few conversions per patch, until all patches are using the new name and finally delete the old name, rather than redefining VIR_ENUM_DECL up front and having to do all the refactoring in a single patch. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org