[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [libvirt] [PATCH] trivial libvirt example code



Jim Meyering wrote:
Dave Allan <dallan redhat com> wrote:
Richard W.M. Jones wrote:
On Fri, Jan 23, 2009 at 02:51:02PM -0500, Dave Allan wrote:
The examples directory doesn't have a trivial example of how to
connect  to a hypervisor, make a few calls, and disconnect, so I
put one  together.  I would appreciate any suggestions on anything
that I've done  wrong as well as suggestions for other fundamental
API calls that should  be illustrated.
Yes, I checked this example code and it is fine.  My only comment
would be on:

+    /* virConnectOpenAuth called here with all default parameters */
+    conn = virConnectOpenAuth(NULL, virConnectAuthPtrDefault, 0);
It might be better to let people connect to a named URI.

Another possibility is to default to the test URI (test:///default)
since that (almost) always exists.
Hi Rich,

Thanks for taking a look at it.  I added a little code to let the user
specify a URI on the command line.  Do you think it is worth
committing?

Hi Dave,

I like your example.
Thanks for preparing it.

Here are some suggestions:

Thanks for the style suggestions--that's one of the reasons I was
sending the code around.

diff --git a/examples/hellolibvirt/hellolibvirt.c b/examples/hellolibvirt/hellolibvirt.c
new file mode 100644
index 0000000..22d3309
--- /dev/null
+++ b/examples/hellolibvirt/hellolibvirt.c
@@ -0,0 +1,151 @@
+/* This file contains trivial example code to connect to the running
+ * hypervisor and gather a few bits of information.  */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <libvirt/libvirt.h>
+
+static int
+showHypervisorInfo(virConnectPtr conn)
+{
+    int ret = 0;
+    unsigned long hvVer, major, minor, release;
+    const char *hvType;
+
+    /* virConnectGetType returns a pointer to a static string, so no
+     * allocation or freeing is necessary; it is possible for the call
+     * to fail if, for example, there is no connection to a
+     * hypervisor, so check what it returns. */
+    hvType = virConnectGetType(conn);
+    if (NULL == hvType) {
+        ret = 1;
+        printf("Failed to get hypervisor type\n");
+        goto out;
+    }
+
+    if (0 != virConnectGetVersion(conn, &hvVer)) {
+        ret = 1;
+        printf("Failed to get hypervisor version\n");
+        goto out;
+    }
+
+    major = hvVer / 1000000;
+    hvVer %= 1000000;
+    minor = hvVer / 1000;
+    release = hvVer % 1000;
+
+    printf("Hypervisor: \"%s\" version: %lu.%lu.%lu\n",
+           hvType,
+           major,
+           minor,
+           release);
+

How about initializing ret = 1 above
and setting ret = 0 here to indicate success?
It's a close call, since that results in removal of
only two "ret = 1" assignments.

In this case, I think that the error cases are very unlikely, so I made the initialization 0, but I agree, it could go either way. I left it as is for now.

+out:
+    return ret;
+}
+
+
+static int
+showDomains(virConnectPtr conn)
+{
+    int ret = 0, i, numNames, numInactiveDomains, numActiveDomains;
+    char **nameList = NULL;
+
+    numActiveDomains = virConnectNumOfDomains(conn);
+    numInactiveDomains = virConnectNumOfDefinedDomains(conn);

It'd be good to handle numInactiveDomains < 0 differently.
Currently it'll probably provoke a failed malloc, below.

Doh--thanks.  I missed that those calls could fail.

+    printf("There are %d active and %d inactive domains\n",
+           numActiveDomains, numInactiveDomains);
+
+    nameList = malloc(sizeof(char *) * (unsigned int)numInactiveDomains);

Using the target variable name rather than the type is a
little more maintainable, in general, so set a good example:
And please drop the cast.  We hate casts, and besides, it's not needed.
       nameList = malloc(sizeof(*nameList) * numInactiveDomains);

Thanks on sizeof(char *) vs. sizeof(*nameList)--fixed.

The cast was there because virConnectNumOfDefinedDomains returns a signed value to allow for returning -1 on error, but malloc expects an unsigned argument. gcc 4.3 -Wconversion complains about this situation [different behavior from gcc < 4.3] I've turned off that warning and removed the cast.

+    if (NULL == nameList) {
+        ret = 1;
+        printf("Could not allocate memory for list of inactive domains\n");
+        goto out;
+    }
+
+    numNames = virConnectListDefinedDomains(conn,
+                                            nameList,
+                                            numInactiveDomains);
+
+    if (-1 == numNames) {
+        ret = 1;
+        printf("Could not get list of defined domains from hypervisor\n");
+        goto out;
+    }
+
+    if (numNames > 0) {
+        printf("Inactive domains:\n");
+    }
+
+    for (i = 0 ; i < numNames ; i++) {
+        printf("  %s\n", *(nameList + i));
+        /* The API documentation doesn't say so, but the names
+         * returned by virConnectListDefinedDomains are strdup'd and
+         * must be freed here.  */
+        free(*(nameList + i));
+    }

Here's another case where you can save a line by initializing
ret=1 up front and setting ret=0 here.

+out:
+    if (NULL != nameList) {
+        free(nameList);

The test for non-NULL-before-free is unnecessary,
since free is guaranteed to handle NULL properly.
So just call free:

       free(nameList);

In fact, if you run "make syntax-check" before making the
suggested change, it should detect and complain about this code.

Removed.  (make syntax-check does not complain, btw)

+    return ret;
+}
+
+
+int
+main(int argc, char *argv[])
+{
+    int ret = 0;
+    virConnectPtr conn = NULL;

The above initialization is unnecessary.

Fixed.

+    char *uri = NULL;

This one can be adjusted, too:

+    printf("Attempting to connect to hypervisor\n");
+
+    if (argc > 0) {
+        uri = argv[1];
+    }

I'd write it as follows,

       char *uri = (argc > 0 ? argv[1] : NULL);

so that it's clear the variable is defined unconditionally.

I tend not to use the ternary operator much, because I've seen it
abused to write really obfuscated code, but you're right, this is a
place where it makes things cleaner.  Done.

In libvirt, it's ok to use C99 declaration-after-stmt.

Good to know.

+    /* virConnectOpenAuth is called here with all default parameters,
+     * except, possibly, the URI of the hypervisor. */
+    conn = virConnectOpenAuth(uri, virConnectAuthPtrDefault, 0);
+
+    if (NULL == conn) {
+        ret = 1;
+        printf("No connection to hypervisor\n");
+        goto out;
+    }
+
+    uri = virConnectGetURI(conn);
+    if (NULL == uri) {
+        ret = 1;
+        printf("Failed to get URI for hypervisor connection\n");
+        goto disconnect;
+    }
+
+    printf("Connected to hypervisor at \"%s\"\n", uri);
+    free(uri);
+
+    if (0 != showHypervisorInfo(conn)) {
+        ret = 1;
+        goto disconnect;
+    }
+
+    if (0 != showDomains(conn)) {
+        ret = 1;
+        goto disconnect;
+    }
+
+disconnect:
+    if (0 != virConnectClose(conn)) {
+        printf("Failed to disconnect from hypervisor\n");
+    } else {
+        printf("Disconnected from hypervisor\n");
+    }

You can save 3 statements by hoisting the initialization of ret=1
and setting ret=0 here.

+out:
+    return ret;
+}

I noticed that you're using the git mirror.  Good!  ;-)
When posting patches, please use "git format-patch".

Will do.

That would have made it easier for me to apply and test
your patches.  As it is, I didn't do either because
"git am FILE" didn't work:

    $ git am dallan.patch
    Applying: trivial libvirt example code
    warning: examples/hellolibvirt/hellolibvirt.c has type 100755, expected 100644
    error: patch failed: examples/hellolibvirt/hellolibvirt.c:97
    error: examples/hellolibvirt/hellolibvirt.c: patch does not apply
    Patch failed at 0001 trivial libvirt example code
    When you have resolved this problem run "git am --resolved".
    If you would prefer to skip this patch, instead run "git am --skip".
    To restore the original branch and stop patching run "git am --abort".

Note the warning about permissions on hellolibvirt.c.
You can correct that by running "chmod a-x hellolibvirt.c".

The permissions problem is strange--it's 644 in my development tree, and
the patch I sent has:
diff --git a/examples/hellolibvirt/hellolibvirt.c
b/examples/hellolibvirt/hellolibvirt.c
new file mode 100644

What would cause git-am to think it was 755?

Here are some contribution guidelines that generally make it
easier for maintainers/committers to deal with contributed patches,
(though some parts are specific to git-managed projects):

    http://git.sv.gnu.org/gitweb/?p=coreutils.git;a=blob;f=HACKING;hb=HEAD

Good stuff.

When I have a patch like this that people have commented on and I've
modified it slightly in response, what's the best way to re-submit it?
When Rich responded, I submitted both the entire patch with the changes
as well as the changes separately.

I'll resend a patch when I've gotten git to squash the history properly to produce something usable from git-format-patch.

Dave


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]