As agreed with phk

bitshift of signed

avoid warning 666 ("Expression with side effects passed to repeated
parameter"): The non-type macros still trigger an warning due to the
expression passed a second time in typeof(). Use the typed macros for
these cases.

avoid error 1058 ("Initializing a non-const reference"): Our type
check (void)(&_va == &_vb) trips for dereferences.

Reviewing flexelint output for the recently added vmin/vmax macros
made apparent that vmin_t/vmax_t basically used typeof((cast)var),
which makes for an unnecessary de-tour via typeof.

We change the min/max macros to explicitly pass the type for the
temporary variables such that typeof() is only needed for vmin/vmax.

Being at it, we also reduce the min/max case to just one take macro
("take the min/max of ...") as they only differed in a single
character, the operator.

Fixes: left operand must be modifiable lvalue: op "="

Fixes warning on 32 bit:

comparison of distinct pointer types ('typeof ((*lp)) *' (aka 'int *') and
'typeof (((VRND_RandomTestable() & 3) + 1)) *' (aka 'long *'))

Minor rewrite for clarity.

VSL lines are truncated, so the return value _is_ sufficiently checked.

Ref 519ead4a

Signed-off-by: Asad Sajjad Ahmed <asadsa@varnish-software.com>

Signed-off-by: Asad Sajjad Ahmed <asadsa@varnish-software.com>

Signed-off-by: Asad Sajjad Ahmed <asadsa@varnish-software.com>

This should make the code easier to read and stop us from being inconsistent.
Signed-off-by: Asad Sajjad Ahmed <asadsa@varnish-software.com>

For the REGEX type, we now peek ahead after the first CSTR token if a '+'
operator follows and, if yes, create the pattern as a concatenation.

Fixes #3726

The disconnect function should not complain when the server is already
closed by an explicit close.

Saw this failure in vtest:

*    top   TEST ../../../../bin/varnishtest/tests/r01918.vtc starting
...
**   s1    === txresp -nolen -bodylen 10
**** s1    txresp|HTTP/1.1 200 OK\r
**** s1    txresp|\r
**** s1    txresp|!"#$%&'()
**   s1    === close
...
**   s1    Waiting for server (4/-1)
**** s1    Closed
***  s1    shutting fd -1
---- s1    Shutdown failed: Bad file descriptor

fixes regression from 9a20cdfd

It needs to be a generic pointer pointer, because it is initialized as
a pointer to a pointer to an arbitrary struct, yet (void **) is can not
be dereferenced.

Right now, the instance info is not used, but this might change.

calling strlen() twice caused an itch

When VJSN is both in libvarnish and libvarnishapi, as in varnishtest,
you would think one of them took precedence, but not so on OSX.

In the shlib (libvarnishapi) text (code) symbols come from the version
in the main binary (varnishtest) which got it from libvarnish, whereas
the data symbols come from the version in libvarnishapi, so we now
hide the data symbols behind code.

Ref 611baec6

... but we can not make any assumption as to why

Fixes #3664

add a vmod function for strftime() formatting in UTC.

As a follow-up issue from #3719, it was noticed that dynamic calls to
built-in SUBs did not work.

This issue was caused by more (struct symbol) members missing
initialization in VCC_New().

In turn, it became apparent that the evaluation context check in
vcc_expr5() as tested in v00020.vtc only worked by accident, and only
for built-in subs because (struct symbol).eval was NULL for built-in
subs.

We fix SUB type expression evaluation with a type specific evaluation
function which allows expression evaluation from explicit SUB contexts
only.

Fixes #3720

vcc_act_call() implements use-before-define semantics. To do so, it
needs to instantiate SUB symbols if they do not exist.

However, it wrongly called VCC_GlobalSymbol() also for existing
symbols, overwriting symbol properties.

In the case of the built-in subs, the symbol's lname (historically)
still is the unescaped literal name, while user defined subs have
their name escaped via VCC_GlobalSymbol() -> VCC_PrintCName().

This made the wrong call to VCC_GlobalSymbol() apparent when a
built-in sub was called with an explicit "call" action.

Besides fixing the VCC_GlobalSymbol() call, we also need to set the
lname and rname attributes for built-in SUB symbols initialized in
VCC_New().

Alternatively, we could also have used VCC_GlobalSymbol() there, but
that would have affected other places where we (still) rely on the
known name scheme of built-in subs, e.h. EmitStruct(). While the name
unifaction was nice in general, I found the necessary changes (look up
SUB symbols) not worth the benefit.

Fixes #3719

Already archived since there are no more occurrences in trunk, but it
could be useful on other branches or third-party libvarnish consumers.

The coccinelle patch also takes care of the recent manual refactoring.

Refs 4ae73a5b
Refs cf00dd2e