[PATCH v2] resolv: Add tests for getaddrinfo returning EAI_AGAIN [BZ #16849]

[Florian Weimer]

* Sergey Kolosov:

> @@ -136,6 +138,7 @@ tests-static += tst-ns_rr_cursor
>  # These tests need libdl.
>  ifeq (yes,$(build-shared))
>  tests += \
> +  tst-getaddrinfo-eai-again-timeout \
>    tst-resolv-ai_idn \
>    tst-resolv-ai_idn-latin1 \
>    tst-resolv-ai_idn-nolibidn2 \
> @@ -280,6 +283,7 @@ $(objpfx)mtrace-tst-resolv-res_ninit.out: $(objpfx)tst-resolv-res_ninit.out
>  
>  $(objpfx)tst-bug18665-tcp: $(objpfx)libresolv.so $(shared-thread-library)
>  $(objpfx)tst-bug18665: $(objpfx)libresolv.so $(shared-thread-library)
> +$(objpfx)tst-getaddrinfo-eai-again-timeout: $(objpfx)libresolv.so $(shared-thread-library)
>  $(objpfx)tst-resolv-ai_idn: $(objpfx)libresolv.so $(shared-thread-library)
>  $(objpfx)tst-resolv-ai_idn-latin1: \
>    $(objpfx)libresolv.so $(shared-thread-library)
> diff --git a/resolv/tst-getaddrinfo-eai-again-timeout.c b/resolv/tst-getaddrinfo-eai-again-timeout.c
> new file mode 100644
> index 0000000000..677bfbe703

> +#include <resolv.h>
> +#include <stdlib.h>
> +#include <string.h>
> +#include <support/check.h>
> +#include <support/check_nss.h>

I'm not sure if check_nss is required?

> +#define TIMEOUT 40

This needs a comment.  Is this large timeout really required?

Maybe we should tweak the resolver configuration so that the test run
more quickly.  I think setting _res.retrans to 1 (measured in seconds,
default is 5) and _res.retry to 1 (retry attempts, default 2) achieves
that.

> diff --git a/resolv/tst-getaddrinfo-eai-again.c b/resolv/tst-getaddrinfo-eai-again.c
> new file mode 100644
> index 0000000000..2a93c1644c
> --- /dev/null
> +++ b/resolv/tst-getaddrinfo-eai-again.c

> +#include <resolv.h>

I think this should be <netdb.h>, where getaddrinfo is declared.

> +#include <stdlib.h>
> +#include <string.h>
> +#include <support/check.h>

> +#include <support/check_nss.h>
> +#include <support/resolv_test.h>
> +#include <support/support.h>

I think those aren't needed anymore.  <stdlib.h> and <string.h> look
superfluous, too.

General direction of this patch looks good.

Thanks,
Florian