Hardening no-STL Builds in C++

.github/workflows/prerelease.yml

@@ -123,7 +123,9 @@ jobs:
             exit 1
           }
       - name: Test C++
-        run: build_artifacts/stringzilla_test_cpp20
+        run: |
+          build_artifacts/stringzilla_test_cpp20
+          build_artifacts/stringzilla_test_avoid_stl_cpp20
       - name: Check stringzilla_bare is built correctly
         run: test -z "$(ldd build_artifacts/libstringzilla_bare.so | grep '.so')"
       - name: Build shared library for Go binding
@@ -264,7 +266,9 @@ jobs:
             exit 1
           }
       - name: Test C++
-        run: build_artifacts/stringzilla_test_cpp20
+        run: |
+          build_artifacts/stringzilla_test_cpp20
+          build_artifacts/stringzilla_test_avoid_stl_cpp20
       - name: Check stringzilla_bare is built correctly
         run: test -z "$(ldd build_artifacts/libstringzilla_bare.so | grep '.so')"
       - name: Build shared library for Go binding
@@ -386,7 +390,9 @@ jobs:
             exit 1
           }
       - name: Test C++
-        run: build_artifacts/stringzilla_test_cpp20
+        run: |
+          build_artifacts/stringzilla_test_cpp20
+          build_artifacts/stringzilla_test_avoid_stl_cpp20
       - name: Check stringzilla_bare is built correctly
         run: test -z "$(ldd build_artifacts/libstringzilla_bare.so | grep '.so')"
 
@@ -482,7 +488,9 @@ jobs:
             exit 1
           }
       - name: Test C++
-        run: build_artifacts/stringzilla_test_cpp20
+        run: |
+          build_artifacts/stringzilla_test_cpp20
+          build_artifacts/stringzilla_test_avoid_stl_cpp20
       - name: Check stringzilla_bare is built correctly
         run: test -z "$(ldd build_artifacts/libstringzilla_bare.so | grep '.so')"
 

CITATION.cff

@@ -6,7 +6,7 @@ license: Apache-2.0
 version: 5.1.2
 repository-code: https://github.com/ashvardanian/StringZilla
 url: https://ashvardanian.com/posts/stringzilla/
-doi: 10.5281/zenodo.21472333
+doi: 10.5281/zenodo.21472332
 abstract: >-
   Portable string-processing library accelerated with SWAR, SIMD, and GPGPU.
   It covers exact and fuzzy search, non-cryptographic hashing and SHA-256 checksums,

CMakeLists.txt

@@ -786,6 +786,11 @@ if (STRINGZILLA_BUILD_TEST)
     define_launcher(stringzilla_test_cpp17 17 "${STRINGZILLA_TARGET_ARCH}" ${STRINGZILLA_TEST_SOURCES})
     define_launcher(stringzilla_test_cpp20 20 "${STRINGZILLA_TARGET_ARCH}" ${STRINGZILLA_TEST_SOURCES})
 
+    # The STL-free build has its own translation unit: the suites above exercise the `std::string` interop that
+    # `SZ_AVOID_STL=1` removes, so they cannot double as its coverage. C++11 pins the floor, C++20 the ceiling.
+    define_launcher(stringzilla_test_avoid_stl_cpp11 11 "${STRINGZILLA_TARGET_ARCH}" test/avoid_stl.cpp)
+    define_launcher(stringzilla_test_avoid_stl_cpp20 20 "${STRINGZILLA_TARGET_ARCH}" test/avoid_stl.cpp)
+
     # Test parallel algorithms separately; link the static archive when the libraries are built. StringZillas requires
     # C++20 as the minimum language standard for concepts, designated initializers, and templated lambdas. We also test
     # C++23 to validate forward compatibility, but only where the toolchain actually advertises support for it — older

README.md

@@ -776,7 +776,7 @@ If StringZilla helps your research or product, please cite it:
 @software{Vardanian_StringZilla,
   author = {Vardanian, Ash},
   title = {{StringZilla: Fast SIMD, SWAR, and GPGPU String Processing}},
-  doi = {10.5281/zenodo.21472333},
+  doi = {10.5281/zenodo.21472332},
   url = {https://github.com/ashvardanian/StringZilla},
   license = {Apache-2.0}
 }

c/stringzillas/stringzillas.cuh

@@ -12,6 +12,7 @@
 
 #include <cstring> // For `std::memcpy`
 
+#include <memory>      // For `std::unique_ptr`
 #include <variant>     // For `std::variant`
 #include <string_view> // For `std::string_view`
 

include/stringzilla/README.md

@@ -1076,7 +1076,11 @@ __`SZ_AVOID_LIBC`__ and __`SZ_OVERRIDE_LIBC`__:
 __`SZ_AVOID_STL`__ and __`SZ_SAFETY_OVER_COMPATIBILITY`__:
 
 > When using the C++ interface one can disable implicit conversions from `std::string` to `sz::string` and back.
-> If not needed, the `<string>` and `<string_view>` headers will be excluded, reducing compilation time.
+> Setting `SZ_AVOID_STL=1` drops STL __interop__, not functionality.
+> The conversions to `std::string` and `std::string_view` disappear, along with the `iterator_category` typedefs that let our iterators satisfy `std::iterator_traits` and the `std::hash` specializations; `sz::string` then defaults to an allocator that refuses every request, so bring your own.
+> Every algorithm and every bounds check stays, including `substr`, `at`, `compare`, `insert`, and `replace`.
+> Failures that would throw `std::out_of_range`, `std::length_error`, or `std::bad_alloc` instead trip `sz_assert_`, which aborts in debug builds and — like the rest of the library's unchecked surface — continues in release builds.
+> The `<iterator>`, `<memory>`, `<string>`, and `<vector>` headers are excluded, which is most of the preprocessing cost: `stringzilla/types.hpp` drops from roughly 58K preprocessed lines to 13K.
 > Moreover, if STL compatibility is a low priority, one can make the API safer by disabling the overloads, which are subjectively error prone.
 
 __`STRINGZILLA_BUILD_SHARED`, `STRINGZILLA_BUILD_TEST`, `STRINGZILLA_BUILD_BENCHMARK`, `STRINGZILLA_TARGET_ARCH`__ for CMake users: