Fix ./configure gssapi and spnego checks under heimdal

[timp87]

configure:36759: checking for working gssapi
.../gssapi/gssapi_krb5.h:130:70: error: unknown type name 'time_t'
...
configure:36835: checking for spnego support
.../gssapi/gssapi_krb5.h:130:70: error: unknown type name 'time_t'

Remove gssapi/gssapi_krb5.h header for gssapi and spnego checks.
Including that header makes checks fail under heimdal.

In Squid v6, this header inclusion was avoided for heimdal. It was
hidden under if USE_HEIMDAL_KRB5 condition. In Squid v7, that
condition no longer exists.

Without the header, the checks work OK for heimdal. Even under MIT krb5,
the checks work fine without this header.

Discovered while porting Squid v7 to FreeBSD.

[timp87]

The error message was:

...
configure:36759: checking for working gssapi
configure:36798: c++ -std=c++17 -o conftest -I/usr/include -O2 -pipe -fstack-protector-strong -isystem /usr/local/include -fno-strict-aliasing  -isystem /usr/local/include  -I/usr/local/include  -I/usr/include -I/usr/include -DBOOL_DEFINED -isystem /usr/local/include   -pthread  -L/usr/local/lib  -L/usr/local/lib -Wl,-R/usr/local/lib conftest.cpp   -L/usr/lib -lkrb5 -lgssapi -lgssapi_krb5  >&5
In file included from conftest.cpp:183:
/usr/include/gssapi/gssapi_krb5.h:130:70: error: unknown type name 'time_t'
  130 | gsskrb5_extract_authtime_from_sec_context(OM_uint32 *, gss_ctx_id_t, time_t *);
      |                                                                      ^
/usr/include/gssapi/gssapi_krb5.h:221:5: error: unknown type name 'int32_t'; did you mean '__int32_t'?
  221 |                                 int32_t *enctypes);
      |                                 ^
/usr/include/sys/_types.h:47:15: note: '__int32_t' declared here
   47 | typedef int                     __int32_t;
      |                                 ^
2 errors generated.


...


configure:36835: checking for spnego support
configure:36887: c++ -std=c++17 -o conftest -I/usr/include -O2 -pipe -fstack-protector-strong -isystem /usr/local/include -fno-strict-aliasing  -isystem /usr/local/include  -I/usr/local/include  -I/usr/include -I/usr/include -DBOOL_DEFINED -isystem /usr/local/include   -pthread  -L/usr/local/lib  -L/usr/local/lib -Wl,-R/usr/local/lib conftest.cpp   -L/usr/lib -lkrb5 -lgssapi -lgssapi_krb5  >&5
In file included from conftest.cpp:184:
/usr/include/gssapi/gssapi_krb5.h:130:70: error: unknown type name 'time_t'
  130 | gsskrb5_extract_authtime_from_sec_context(OM_uint32 *, gss_ctx_id_t, time_t *);
      |                                                                      ^
/usr/include/gssapi/gssapi_krb5.h:221:5: error: unknown type name 'int32_t'; did you mean '__int32_t'?
  221 |                                 int32_t *enctypes);
      |                                 ^
/usr/include/sys/_types.h:47:15: note: '__int32_t' declared here
   47 | typedef int                     __int32_t;
      |                                 ^
2 errors generated.
...

[yadij]

It looks to me like these tests have correctly identified a problem with the gssapi/gssapi_krb5.h files. They are included in the same manner by Squid code (i.e. without a Heimdal conditional).

IMO, correct fix would be to add #include <ctime> to these tests.

[rousskov]

It looks to me like these tests have correctly identified a problem with the gssapi/gssapi_krb5.h files. They are included in the same manner by Squid code (i.e. without a Heimdal conditional).

I agree that ./configure checks included gssapi/gssapi_krb5.h incorrectly. We need to decide whether they should include that header (correctly).

IMO, correct fix would be to add #include <ctime> to these tests.

If gssapi/gssapi_krb5.h is never needed for the tests in question, then that header should not be included at all (i.e. the current PR changes are correct). It may not be needed because because gss_create_empty_oid_set() and gss_indicate_mechs() are declared in/via gssapi/gssapi.h instead. However, I do not know whether that holds for all build environments, and whether some other test code needs that header to be included.

I adjusted PR title and description, primarily to quote the corresponding error messages.

[timp87]

@yadij BTW I just tested adding #include <ctime> next to #include <gssapi/gssapi_krb5.h> in the gssapi and spnego checks.
The error is still the same. Doesn't matter if ctime is before or after gssapi_krb5.h.

[rousskov]

@yadij BTW I just tested adding #include <ctime> next to #include <gssapi/gssapi_krb5.h> in the gssapi and spnego checks. The error is still the same. Doesn't matter if ctime is before or after gssapi_krb5.h.

That makes sense because <ctime> does not declare time_t. It declares std::time_t. time_t is declared in <time.h>.

In my tests, <ctime> does include <time.h> but that inclusion could be platform-specific. I cannot easily test on FreeBSD right now.

[timp87]

@rousskov right! Including <time.h> doesn't help as well, but <sys/types.h> does.

[rousskov]

Including <time.h> doesn't help as well, but <sys/types.h> does.

It feels odd to me that <time.h> does not define time_t, but this is not my area of expertise. If I had to guess, I would bet that there is more than one type missing, and <sys/types.h> provides the other missing type(s), fixing the test. Checking the exact compiler error messages could confirm or reject that speculation.

[timp87]

This PR has S-waiting-for-author label.
Is something missing from me?

[rousskov]

This PR has S-waiting-for-author label. Is something missing from me?

Based on the discussion so far, including the failed attempt to add ctime (as suggested) and the proposed decision logic that got author support and matches current PR state, I think it is time to request another review. The ball is on Amos side now.