* 'intrusive_ptr' of https://github.com/MaxKellermann/zeek: (32 commits)
Scope: store IntrusivePtr in `local`
Scope: pass IntrusivePtr to AddInit()
DNS_Mgr: use class IntrusivePtr
Scope: use class IntrusivePtr
Attr: use class IntrusivePtr
Expr: check_and_promote_expr() returns IntrusivePtr
Frame: use class IntrusivePtr
Val: RecordVal::LookupWithDefault() returns IntrusivePtr
Type: RecordType::FieldDefault() returns IntrusivePtr
Val: TableVal::Delete() returns IntrusivePtr
Type: base_type() returns IntrusivePtr
Type: init_type() returns IntrusivePtr
Type: merge_types() returns IntrusivePtr
Type: use class IntrusivePtr in VectorType
Type: use class IntrusivePtr in EnumType
Type: use class IntrusivePtr in FileType
Type: use class IntrusivePtr in TypeDecl
Type: make TypeDecl `final` and the dtor non-`virtual`
Type: use class IntrusivePtr in TypeType
Type: use class IntrusivePtr in FuncType
...
This also now installs the full rapidjson include/ tree in order to
allow external plugins to properly build if they include
threading/formatters/JSON.h
* origin/topic/jsiwek/32bit-compat:
Improve formatting of doubles that are close to integers
Improve HTTP version number comparisons
Add a 32-bit task to Cirrus CI config
Replace va_list fmt() overload with vfmt()
Format tables indexed by patterns consistently across 32-bit/64-bit
Format interval values consistently across 32-bit/64-bit platforms
Using an overload that takes a va_list argument potentially causes
accidental misuse on platforms (e.g. 32-bit) where va_list is
implemented as a type that may collide with commonly-used argument
types.
For example:
char* c = copy_string("hi");
fmt("%s", (const char*)c);
fmt("%s", c);
The first fmt() call correctly goes through fmt(const char*, ...) first,
but the second mistakenly goes through fmt(const char*, va_list) first
because variadic function overloads have lower priority during overload
resolution and va_list on a 32-bit system happens to be defined as a
pointer type that can match with "char*" but not "const char*".
The Zeek code base has very inconsistent #includes. Many sources
included a few headers, and those headers included other headers, and
in the end, nearly everything is included everywhere, so missing
#includes were never noticed. Another side effect was a lot of header
bloat which slows down the build.
First step to fix it: in each source file, its own header should be
included first to verify that each header's includes are correct, and
none is missing.
After adding the missing #includes, I replaced lots of #includes
inside headers with class forward declarations. In most headers,
object pointers are never referenced, so declaring the function
prototypes with forward-declared classes is just fine.
This patch speeds up the build by 19%, because each compilation unit
gets smaller. Here are the "time" numbers for a fresh build (with a
warm page cache but without ccache):
Before this patch:
3144.94user 161.63system 3:02.87elapsed 1808%CPU (0avgtext+0avgdata 2168608maxresident)k
760inputs+12008400outputs (1511major+57747204minor)pagefaults 0swaps
After this patch:
2565.17user 141.83system 2:25.46elapsed 1860%CPU (0avgtext+0avgdata 1489076maxresident)k
72576inputs+9130920outputs (1667major+49400430minor)pagefaults 0swaps
This would mistakenly have the Stem process kill itself due to giving
PID 0 as argument to kill() where it really was being used to mean "that
node does not currently have any live process associated with it" and so
can just be removed without trying to kill/reap.
