firmware: Fix unused-var and type mismatch error in recent gcc (7+). #448
Conversation
| @@ -70,6 +70,8 @@ void pll_config_for_clock(int freq) | |||
| * Reproducible both with DRP and initial reconfiguration. | |||
| * Until this spartan6 weirdness is sorted out, just stick to 20x. | |||
| */ | |||
|
|
|||
| (void) pll_config_10x[0]; // Dummy access to suppress to suppress -Werror=unused-const-variable. | |||
There was a problem hiding this comment.
I think we want to use something like https://stackoverflow.com/questions/4851075/universally-compiler-independent-way-of-implementing-an-unused-macro-in-c-c
There was a problem hiding this comment.
There was a problem hiding this comment.
Clearly, there is no universal solution :P. the sizeof solution causes warnings on newer compilers.
I think unused vars are infrequent enough that I prefer my solution with works just fine for this single use case. That said, if you really want a macro, __attribute__(unused) should be fine, since gcc/clang is currently the only compiler for all CPUs we support. But I would prefer to gate it behind a macro like the second answer (just in case).
What header in HDMI2USB/litex-buildenv should such a macro go?
|
@cr1901 Any idea why we are not seeing this issue on Travis now I have rolled the latest GCC? |
|
No, I haven't a clue. Perhaps GCC's warnings got less pessimistic? |
|
It seems like this pull request has already been merged. I'm going to close it. |
When compiling the HDMI2USB firmware w/ recent gcc versions (I first noticed them in 7+), the build would error out in
pll.candci.c. These are my proposed fixes.The
pll.cworkaround(void) pll_config10x[0];is a no-op that makes gcc (and probably other C compilers) treat a variable as used.-Werror=unused-const-variableis enabled, and since it's a conditional compile for now that determines whetherpll_config10x[0]actually gets used, I think this is a better solution until the S6 issues get worked out.The
ci.cfix was attempting to compare achar *to acharliteral, and I believe that to be a genuine bug. :)