win32/perlhost.h: revert changes to SETUPEXCHANGE

[mauke]

This was broken by 48bda52, which removed the * before xptr. Without *, we set our iptr (a class member) to a bogus pointer, and the assignment to xptr (which is a function parameter and hence a local variable) is effectively dead code.

---

• This set of changes does not require a perldelta entry.

[bulk88]

I have no personal opinion (yes/no wise) on this PR/this patch, and have no opinion on what to with macro SETUPEXCHANGE.

based on an IRC talk. Me, and maybe a 2nd person think that maybe-public API front end fnperl_alloc_override() which is a 5.005-5.41 quasi-public API Win32 exported symbol from perl5XX.dll , along with perl_alloc_override()'s static declared backend, that is called CPerlHost::CPerlHost(const struct IPerlMem** ppMem, const struct IPerlMem** ppMemShared, const struct IPerlMem** ppMemParse, const struct IPerlEnv** ppEnv, const struct IPerlStdIO** ppStdIO, const struct IPerlLIO** ppLIO, const struct IPerlDir** ppDir, const struct IPerlSock** ppSock, const struct IPerlProc** ppProc) are dead or abandoned code that has no consumers/end users, no tests in core or CPAN, and its machine code has never been executed inside a CPU since the day it was written.

P5P's psuedo-fork feature doesn't call this func, neither does CPAN's threads.pm.tar.gz and CPAN's threads::shared.pm.tar.gz.

ZERO HITS on p5p ML (help the website I used before covid shutdown and post covid Google refuses to index nntp.perl.org)

https://groups.google.com/search?q=perl_alloc_override

ZERO HITS gmcpan

https://grep.metacpan.org/search?size=20&_bb=913286821&q=perl_alloc_override&qd=&qft=&qifl=

TonyC has a long time suspicion ActiveState's Perl for MS IIS product uses perl_alloc_override(). I decided to research that, and I found by accident a perlis.dll on my old box. My file was built against/C links against perl510.dll.

A printable ascii string dump from my perlis.dll. Yes thats it for strings. perlis.dll is a very small lightweight library.

.rdata:281A46E0 KERNEL32.dll
.rdata:281A4772 ADVAPI32.dll
.rdata:281A4896 MSVCRT.dll
.rdata:281A493C PerlIS.dll
.rdata:281A4947 GetExtensionVersion
.rdata:281A495B HttpExtensionProc
.data:281A5024  PerlIS.dll was unable to open PerlIS-Err.log
.data:281A5054  PerlIS.dll
.data:281A5060  PerlIS-Err.log
.data:281A5070  \r\n*** error message at: %4d/%02d/%02d %02d:%02d:%02d\r\n
.data:281A50A8  \r\n*** [%u] error message at: %4d/%02d/%02d %02d:%02d:%02d\r\n
.data:281A50E4  \r\n*** '%s' error message at: %4d/%02d/%02d %02d:%02d:%02d\r\n
.data:281A5120  \r\n*** '%s' [%u] error message at: %4d/%02d/%02d %02d:%02d:%02d\r\n
.data:281A51B4  Date: %s, %02d %s %d %02d:%02d:%02d GMT\r\n
.data:281A522C  SERVER_SOFTWARE
.data:281A523C  Server:
.data:281A5248   200 OK\r\n
.data:281A5254   302 Moved Temporarily\r\n
.data:281A5270  SERVER_PROTOCOL
.data:281A5284  ocation:
.data:281A5290  tatus: 302
.data:281A529C  ALL_HTTP
.data:281A52A8  CONTENT_TYPE
.data:281A52BC  CONTENT_LENGTH
.data:281A52CC  PATH_TRANSLATED
.data:281A52DC  PATH_INFO
.data:281A52E8  QUERY_STRING
.data:281A52F8  REQUEST_METHOD
.data:281A5308  <br>
.data:281A5310   200 OK\nContent-type: text/html\n\n
.data:281A54FC  PerlIS does not support shelling (Execl)
.data:281A5528  PerlIS does not support shelling (Execv)
.data:281A5554  PerlIS does not support shelling (Execvp)
.data:281A5580  fork() is not implemented in PerlIS
.data:281A55A4  Error: Unable to allocate memory
.data:281A55CC  Error: Unable to construct data structures
.data:281A5604  Error: Parse exception\n
.data:281A561C  Error: Runtime exception\n
.data:281A5638  :unix
.data:281A5640  DynaLoader::boot_DynaLoader
.data:281A565C  ..\\Common\\PerlHost\\ExtHost.cpp
.data:281A567C  SetSVFromVariantEx
.data:281A5690  SetSVFromVariant
.data:281A56A4  SetVariantFromSV
.data:281A56B8  CreatePerlObject
.data:281A56CC  ole.dll
.data:281A56D4  Perl for Win32
.data:281A56E4  SCRIPT_NAME
.data:281A56F4  EnableImpersonate
.data:281A5708  EnableEventLogMsgs
.data:281A571C  EnableDebugOutput
.data:281A5730  EnableCGIHeader
.data:281A5740  SOFTWARE\\ActiveState\\PerlIS
.data:281A575C  Perl_PerlIO_stderr
.data:281A5770  Perl_PerlIO_stdout
.data:281A5784  Perl_PerlIO_stdin
.data:281A5798  Perl_PerlIO_flush
.data:281A57AC  Perl_PerlIO_close
.data:281A57C0  PerlIO_binmode
.data:281A57D0  PerlIO_fdopen
.data:281A57E0  PerlIO_findFILE
.data:281A57F0  Perl_do_spawn_nowait
.data:281A5808  Perl_set_context
.data:281A581C  Perl_get_context
.data:281A5830  Perl_newXS
.data:281A583C  Perl_safesysfree
.data:281A5850  Perl_Gcurinterp_ptr
.data:281A5864  Perl_Iexit_flags_ptr
.data:281A587C  Perl_Iperl_destruct_level_ptr
.data:281A589C  Perl_Isys_intern_ptr
.data:281A58B4  Perl_IDir_ptr
.data:281A58C4  Perl_croak_nocontext
.data:281A58DC  Perl_Itop_env_ptr
.data:281A58F0  Perl_safesysrealloc
.data:281A5904  Perl_safesysmalloc
.data:281A5918  perl_destruct
.data:281A5928  perl_run
.data:281A5934  perl_parse
.data:281A5940  perl_alloc_override
.data:281A5954  perl_construct
.data:281A5964  perl_free
.data:281A5970  boot_DynaLoader
.data:281A5980  Perl_sys_term
.data:281A5990  Perl_sys_init
.data:281A59A0  perl
.data:281A59A8  Perl510.dll
.data:281A59B4  PerlMsg
.data:281A59BC  No script was supplied to PerlIS.dll\n
.data:281A59E8  PerlCreate failed!!!!
.data:281A5A00  INTERNAL ERROR: %s\n
.data:281A5A14  PerlParse did not exit clean!!!!
.data:281A5A38  PerlXS=PerlIS
.data:281A5A48  PerlHost=PerlIS
.data:281A5A58  Content-type: text/html\n\n
.data:281A5A74  '%s' script produced no output\r\n
.data:281A5AA8  .?AVtype_info@@

My uneducated wild guess says, Perl-4-IIS uses a pre-fork architecture built ontop of IIS's random and self scaling thread pool and IIS's event loop and its idle loop. Also PerlIIS has to disable libc-level stdio (shared hosting/anti malware/typical daily black hat problems) .

Also, very critically, PerlIIS must C level trap and intercept and prevent libperl5xx.dll from calling libc exit(). PP keyword level, XS level, C level, doesn't matter, MSCRT's exit() and _exit() have to be trapped or say bye bye to the HTTP server process.

Don't know if this is true since perlis.dll is closed source, but maybe perlis.dll hooks all 3 WinPerl memory pools, PerlMem, PerlMemShared, PerlMemParse, and replaces them with a magic-IIS-native-MS-COM/OLE allocator, so Perl PP level scalar strings buffers can either

• 1 be cleaned up after every http request and after every psudo-pre-fork.

• 2 Clean up memory leaks in PerlMemShare and PerlMemParse.

Many many years ago I real or someone told me OP slab API, which allocates in 4KB units, never ever will release memory back to libc/OS kernel/Ring 0. Either an actual leak in Perl Core b/c CV*--; RC==0; freeing doesn't free the OP structs b/c lack of atomics or lack of a RC struct field in the header of an op blab. It could also be a performance optimization, like how SV head and body arenas never shrink. And the PL_stack_sp never shrinks. Or the OP slabs are being correctly tracked and freed, with proven later reuse, but "freed OP slabs" get placed into a P5P controlled free memory pool. But a Perl 5 interp controlled free memory pool is useless for preventing memory leaks in top and ps 😁

So maybe Perl-4-IIS also hooks and occasionally wipes the PerlMemParse and PerlMemShare pools. P5P Public API official legal answer would be execute perl_sys_term(); perl_sys_init() on a timer/automatic trigger. Real world conditions would say thats unfeasible in a production grade Perl process and its time to hack the interp's private API.

Other thing, hypothetical, PerlIIS is definitely a pre-docker product. Maybe it was optimized for shared hosting with a Windows API chroot jail, 2000 unique customer perl back end websites, hosted by one iis.exe each website getting 2 hits a week. Zero chance of wining a disk OS kernel cache hit of a .pl/.pm block device file, and zero chance of a winning hit inside same address space/same OS process, at any of these SW layers, kernel/MS CRT/P5P interp/AS Perl-4-IIS/MS IIS daemon/CPAN XS/Pure Perl, So maybe the PerlMemParse and PerlMemShare pools do leak really badly if u execute

    push @INC, ".";
    $i = 10000;
    while(--$i) {
         if( ! fork_fake()) {
               do './warnings.pm
         }
     }

Since Perl-4-IIS NEVER links with symbol Perl_call_sv, thats giving me a suspicion its doing a pre-fork arch at best, and at worst, launches a totally isolated my_perl instance with properl WinNT ACLs for each customer account inside 1 huge multi tenant iis.exe process. Not every web hosting platform is Heroku or Amazon or Azure or an identical twin of those 3.

So ^^^^^ is design considerations on what embedders really need feature wise if life was perfect.

@bulk88 will always advocate for, or write patches to actually implement, swapping out WinPerl's memory allocator system.

At runtime. With -O2 stable WinPerl. Using only CPAN XS. Default on/perma on in all build configs of WinPerl.
Absolutely no recompiling the interp binary. No -DDEBUGGING, no other special P5P secret build flags. Yes I know about the very nice leaking SV* feature, and I know about the very nice Posix+WinPerl malloc/Newx leak tracking feature. I've used them. They are very nice. But I hate spending 7 hours at 50-100% CPU to compile -DDEBUGGING with MSVC 2022 to reach gmake test-prep stage.

I'll probably need to open another ticket, with some benchmarks, stack traces from inside cl.exe/link.exe and demand that someone go inside toke.c, regexec.c, regcomp.c with a chainsaw. My MSVC 2022 cl.exe/link.exe processes are actually SEGVing in virtual address space during the normal -O1 compiling process while writing regexec.obj to disk. Long long bug ticket I don't want to write. My MSVC 2022 last summer also created a 140MB .dll disk file, from my 1 and only C function in my .c that I fed into MSVC. Strawberry GCC created a 3.8MB .dll disk file with exact same .c file as input. The C function was generated by ExtUtils::Constant, for a data set of ummmm, lets leave it at, the entire national license plate database of a small european country. @bulk88 needs to stop writing perl and invest in learning SQL.