<!DOCTYPE html>
<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body>
    <p>Thanks for hint. I have created ticket:</p>
    <p><a class="moz-txt-link-freetext" href="https://gitlab.com/freepascal.org/fpc/source/-/work_items/41826">https://gitlab.com/freepascal.org/fpc/source/-/work_items/41826</a></p>
    <div class="moz-cite-prefix">W dniu 22.07.2026 o 20:02, n7800 pisze:<br>
    </div>
    <blockquote type="cite"
      cite="mid:1784743354.941298159@f328.i.mail.ru">
      <meta http-equiv="content-type" content="text/html; charset=UTF-8">
      <div class="cl-0u1zw232om">
        <div>
          <div>Perhaps it would be better to post this on the <a
href="https://gitlab.com/freepascal.org/fpc/source/-/work_items"
              moz-do-not-send="true">FPC bug tracker</a>?</div>
          <div> </div>
          <div>Or on the "fpc-devel" mailing list.</div>
        </div>
        <br>
        <div class="mail-quote-collapse">
          <blockquote
style="border-left:1px solid #0857A6;margin:10px;padding:0 0 0 10px">
            <div>
              <div id="">
                <div class="cl-gwbfz3ahak">
                  <div
class="js-helper_mr_css_attr js-readmsg-msg_mr_css_attr">
                    <div id="style_17847294010710600985_mr_css_attr">
                      <div
                        id="style_17847294010710600985_BODY_mr_css_attr">
                        <p>Hi,</p>
                        <p><span lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">I suspect
                                that my first message was blocked
                                because I attached ODT document. Please
                                ignore duplicate if it show up.</span></span></span></p>
                        <p>I mentioned about this issue some time ago in
                          the topic about PasFLTK bindings which I made.
                          I decided to make dedicated thread because it
                          seems to be interesting.</p>
                        <p>So. I'm developing PasFLTK bindings for CFLTK
                          which is flatten C binding for major FLTK UI
                          framework written in C++. PasFLTK has two
                          modes, shared library which work perfectly
                          fine and experimental static linking.</p>
                        <p>Static linking has issues which I'm <span
                            lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">consistently</span></span></span> fixing.
                          Major problem is that C++ classes are not
                          initialized in proper way. Problem is not a
                          matter of "Static Initialization Order Fiasco"
                          because CFLTK/FLTK is designed to be static
                          linking for default (and that is how other
                          language bindings are using it). Problem is
                          how FPC LD linker prepare "startup" routines.
                          For example:</p>
                        <p>Class Fl_File_Chooser2.cxx has this public
                          variable:</p>
                        <p>const char     
                          *Fl_File_Chooser::filesystems_label =
                          Fl::system_driver()->filesystems_label();</p>
                        <p>System_driver is not initialized yet so
                          filesystems_label become null which result
                          with access violations. Since I'm linux user I
                          started from this platform, working on this
                          with Claude AI (which amazed me how deeply he
                          can analyze binaries on low level). On linux,
                          fix was small and easy (for Claude, not for me
                          of course). LD linker prepare wrong init
                          structure and doesn't call the
                          __init_array_start in linux ELF standard. So
                          patch looks like that:<br>
                           </p>
                        <p>type<br>
                            TCtorProc = procedure(); cdecl;<br>
                            PCtorProc = ^TCtorProc;<br>
                          <br>
                          var<br>
                            __init_array_start: Byte; external name
                          '__init_array_start';<br>
                            __init_array_end: Byte; external name
                          '__init_array_end';<br>
                          <br>
                          procedure RunCppGlobalConstructorsLinux;<br>
                          var<br>
                            p, pend: PCtorProc;<br>
                          begin<br>
                            p := PCtorProc(@__init_array_start);<br>
                            pend := PCtorProc(@__init_array_end);<br>
                            while p < pend do<br>
                            begin<br>
                              if Assigned(p^) then<br>
                                p^();<br>
                              Inc(p);<br>
                            end;<br>
                          end; <br>
                          <br>
                          That small piece of code fixed everything and
                          static linking work like a charm on linux. I
                          have project (which I publish soon) totally
                          written as static linking and didn't notice
                          any new issues so far.</p>
                        <p>Now windows version. Here complication start.
                          First, build-in FPC linker is useless and
                          throw error:</p>
                        <p>Error: Failed reading coff file, invalid
                          section index while reading
                          C:\Users\vboxuser\Documents\CFLTKlibs\libcfltk.a(cfl_window.cpp.obj)<br>
                          <br>
                          I had to use -Xe flag which switch to external
                          ld.exe linker. But then I had DWARF errors
                          because LD support only up to 4 version and my
                          MinGW toolchain (MSYS2 on windows) is using
                          ver. 5. I had to call `strip --strip-debug` on
                          all *.a files (including some MinGW - e.g
                          libstd++.a).</p>
                        <p>Now, when binding are compiling fine, I fall
                          to the same problem with initialization as on
                          linux. I spent few evenings with Claude on
                          this. We compared objdump -h and extracted
                          objects on exactly the same demos written line
                          by line one using C (CFLTK) and second Free
                          Pascal. Since linux use ELF we get
                          __init_array_start, on windows PE - it looks
                          harder. LD.exe <span lang="en"
                            class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">delivered
                                with FPC is quite old and doesn't use
                                modern standard. Demo written in C and
                                linked by LD from toolchain has correct
                                .ctors and __CTOR_LIST__ which FPC
                                ld.exe is missing (in fact, filling it
                                incorrectly). Tried using
                                -FD"C:\msys64\mingw64\bin" which load
                                ld.exe from MinGW toolchain but this
                                result with dozens errors:</span></span></span></p>
                        <p><span lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">C:\msys64\mingw64\bin\ld.exe:
C:\programowanie\lazarus\fpc\3.2.2\units\x86_64-win64\rtl\sysutils.o:
                                illegal relocation type 0 at address 0<br>
                                C:\msys64\mingw64\bin\ld.exe:
                                C:\programowanie\lazarus\fpc\3.2.2\units\x86_64-win64\rtl\math.o:
                                illegal relocation type 0 at address 0</span></span></span></p>
                        <p><span lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">Claude also
                                compared rust bindings for CFLTK because
                                it also use static linking. They had
                                also problems and are stick to the
                                specific toolchain. But there problem
                                was easier to fix because Rust linker
                                correctly linking crt2.o and has own
                                guardians rsbegin.o and rsend.o (</span></span></span><span
                            lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">equivalent
                                of </span></span></span>crtbegin.o
                          and crtend.o) which are missing in FPC ld
                          linker.</p>
                        <p>Finally I get it working with Claude but this
                          require reorganize *.a and C++ objects with
                          --globalize-symbol and this solution is not
                          acceptable for me because it require <span
                            lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">engagement
                                everytime when something changed in
                                CFLT/FLTK source or even in MinGW.</span></span></span></p>
                        <p><span lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">I'll be not
                                writting anymore and just attach summary
                                of our work which I asked Claude to
                                write in short. It has more technical
                                details. To sum up. Is there any plans
                                to upgrade LD linker or exists any alpha
                                / beta of upcoming release? </span></span></span><span
                            lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">The
                                difference between FPC ld and GCC ld is
                                almost 10 years. </span></span></span><span
                            lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">I read
                                somewhere that FPC is not designed to
                                link C++ (only C) but as we can see, it
                                could be just small step to make it
                                possible and that open huge door. </span></span></span></p>
                        <p><span lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">Below is
                                Claude's summary.</span></span></span></p>
                        <p><span lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">Regards,
                                Dibo.</span></span></span></p>
                        <p><span lang="en" class="HwtZe_mr_css_attr"><span
class="jCAhz_mr_css_attr ChMk0b_mr_css_attr"><span
                                class="ryNqvb_mr_css_attr">Environment:<br>
                                <br>
                                    Free Pascal Compiler 3.2.2, Windows
                                target x86_64-win64<br>
                                <br>
                                    Bundled linker:
                                C:\programowanie\lazarus\fpc\3.2.2\bin\x86_64-win64\ld.exe
                                — GNU ld (GNU Binutils) 2.28<br>
                                <br>
                                    Comparison linker: ld.exe from MSYS2
                                mingw-w64-x86_64-binutils package,
                                located at C:\msys64\mingw64\bin\ld.exe
                                — GNU ld (GNU Binutils) 2.46.1<br>
                                <br>
                                    GCC/G++ used to build FLTK, CFLTK,
                                and the comparison C++ demo: 16.1.0
                                (MSYS2 mingw-w64-x86_64-gcc, reported as
                                "GCC 16.1.0"<br>
                                <br>
                                    FLTK version: 1.4.5, CFLTK version:
                                1.5.23<br>
                                <br>
                                Root cause identified: FPC 3.2.2's
                                bundled ld.exe fails to correctly merge
                                .ctors input sections (containing C++
                                global constructor pointers) from object
                                files compiled by a modern GCC (16.1.0,
                                MinGW-w64/MSYS2) into the final
                                __CTOR_LIST__ array — even though the
                                linker script (identical KEEP(*(.ctors))
                                rule, confirmed via ld.exe --verbose on
                                both linkers) is present in both the old
                                bundled ld.exe and a current MSYS2
                                ld.exe.<br>
                                <br>
                                Evidence:<br>
                                <br>
                                    A raw hex dump of __CTOR_LIST__ in
                                an FPC-linked executable (with 10 real
                                global constructors from a
                                statically-linked FLTK/CFLTK library
                                present and confirmed via nm) shows:
                                sentinel (-1) immediately followed by
                                terminator (0) — zero constructor
                                pointers, despite the constructor
                                functions (_GLOBAL__sub_I_*) physically
                                existing in the binary.<br>
                                <br>
                                    The same source compiled and linked
                                natively with g++.exe/current MSYS2
                                ld.exe produces a fully populated
                                __CTOR_LIST__ with all 10 pointers
                                correctly resolved to the matching
                                _GLOBAL__sub_I_* addresses.<br>
                                <br>
                                    Attempting to relink the FPC-built
                                binary with the current MSYS2 ld.exe
                                (via -FD<path>) fails with illegal
                                relocation type 0 / invalid section
                                index errors on FPC's own RTL object
                                files (sysutils.o, math.o) — indicating
                                FPC's internal object emitter and modern
                                binutils have diverged in mutually
                                incompatible ways. rtl\sysutils.o:
                                illegal relocation type 0 at address 0
                                etc<br>
                                <br>
                                    The input .o files from the affected
                                FLTK library use COMDAT sections
                                (LINK_ONCE_DISCARD) extensively
                                (confirmed via objdump -h), which may be
                                related to how the old ld.exe mishandles
                                their .ctors content during merging —
                                this is hypothesis<br>
                                <br>
                                    Consequently, statically linking
                                modern C++ libraries (FLTK/CFLTK, built
                                with current MinGW-w64/GCC) into an
                                FPC/Lazarus Windows executable silently
                                drops all global C++ constructors,
                                causing use of uninitialized static
                                objects downstream (observed as
                                intermittent access violations depending
                                on which code paths touch affected
                                statics).<br>
                                <br>
                                Workaround currently used: manually
                                enumerating _GLOBAL__sub_I_* symbols via
                                nm on the linked (even if crashing)
                                executable, globalizing their linkage
                                scope via objcopy --globalize-symbol
                                (they are emitted as local/t symbols),
                                and generating a small Pascal unit that
                                calls them all explicitly at program
                                startup. This resolves some but not all
                                downstream issues, suggesting missing
                                constructor invocation is necessary but
                                possibly not sufficient — full
                                mainCRTStartup context
                                (TLS/locale/guard-variable state) may
                                also matter for some lazily-initialized
                                C++ singletons. - this is last
                                hypothesis<br>
                                <br>
                                Question for the list: Is there a
                                newer/beta ld.exe (or plan to update the
                                bundled binutils) for the Windows FPC
                                toolchain that correctly handles
                                .ctors/COMDAT sections as emitted by
                                current GCC, without breaking
                                compatibility with FPC's own object file
                                emitter?</span></span></span></p>
                        <p> </p>
                      </div>
                      <div>_______________________________________________<br>
                        fpc-pascal maillist - <a
                          href="mailto:fpc-pascal@lists.freepascal.org"
                          moz-do-not-send="true"
                          class="moz-txt-link-freetext">fpc-pascal@lists.freepascal.org</a><br>
                        <a
href="https://lists.freepascal.org/cgi-bin/mailman/listinfo/fpc-pascal"
                          moz-do-not-send="true"
                          class="moz-txt-link-freetext">https://lists.freepascal.org/cgi-bin/mailman/listinfo/fpc-pascal</a></div>
                    </div>
                  </div>
                </div>
              </div>
            </div>
          </blockquote>
        </div>
      </div>
    </blockquote>
  </body>
</html>