Fix memory leaks due to early returns in opus_compare.c and opus_demo.c about opus

[jmvalin]

We're aware of the opus_compare.c issue and it's a WONTFIX. That file is part of the standard for testing Opus compliance so we're not changing it at all unless something really bad gets discovered. When it comes to opus_demo, it would be good to have a less repetitive patch (maybe using goto?) to merge.

[ddkilzer]

Seems odd to leave known issues unfixed in test code, but okay.

I’m not allowed to post patches to other projects without internal review, so this bug report documents the issue. You’re free to take the existing patch and use it, modify it or fix it a different way. I’m not a huge fan of goto statements, anyway. :)

[jmvalin]

About opus_compare.c, let's just say that it's as much of a contract and political statement as it is software, so any change to it makes some people nervous. As for opus_demo.c, can you have a look at this goto-based patch I made and give it a try?
https://jmvalin.ca/misc_stuff/fix_opus_demo_leaks.patch
As much as I also hate gotos in general, I think this is the one use case, where it's cleaner than alternatives. I believe it also addresses a case that your patch missed (exiting on opus_decoder_create() would fail to destroy the encoder)

[ddkilzer]

Okay, will try to do a code review here.

-static void check_encoder_option(int decode_only, const char *opt)
-{
-   if (decode_only)
-   {
-      fprintf(stderr, "option %s is only for encoding\n", opt);
-      exit(EXIT_FAILURE);
-   }
-}
+#define check_encoder_option(decode_only, opt) do {if (decode_only) {fprintf(stderr, "option %s is only for encoding\n", opt); goto failure;}} while(0)

I would find the macro much easier to read if it were formatted like regular code with line continuation backslashes at the end of each line:

+#define check_encoder_option(decode_only, opt) \
+do { \
+    if (decode_only) { \
+        fprintf(stderr, "option %s is only for encoding\n", opt); \
+        goto failure; \
+    } \
+} while(0)

Also learned that calling free(NULL) is okay. (Didn't realize that!)

Patch looks good to me other than the readability of check_encoder_option() macro.