FIX: SLEEF migration to 4.0

.github/workflows/big_endian.yml

@@ -22,7 +22,7 @@ permissions:
 
 jobs:
   big_endian_tests:
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-24.04
     continue-on-error: true
     strategy:
       fail-fast: false

quaddtype/meson.build

@@ -25,8 +25,9 @@ if sleef_dep.found() and sleef_dep.type_name() != 'internal'
   # SLEEF found system-wide - verify quad-precision support
   cpp = meson.get_compiler('cpp')
   sleefquad_lib = cpp.find_library('sleefquad', required: false)
+  tlfloat_lib = cpp.find_library('tlfloat', required: false)
 
-  if sleefquad_lib.found()
+  if sleefquad_lib.found() and tlfloat_lib.found()
     sleefquad_test_code = '''
     #include <sleefquad.h>
 
@@ -40,14 +41,17 @@ if sleef_dep.found() and sleef_dep.type_name() != 'internal'
     # this should compile and link
     quad_works = cpp.links(
       sleefquad_test_code,
-      dependencies: [sleef_dep, sleefquad_lib],
+      dependencies: [sleef_dep, sleefquad_lib, tlfloat_lib],
       name: 'SLEEF quad-precision support'
     )
 
     if quad_works
       sleefquad_dep = declare_dependency(
         dependencies: [sleef_dep, sleefquad_lib]
       )
+      tlfloat_dep = declare_dependency(
+        dependencies: [tlfloat_lib]
+      )
       use_system_sleef = true
     else
       fallback_reason = 'System-wide SLEEF installation found but a test for quad precision support failed.'
@@ -65,6 +69,7 @@ else
   sleef_subproj = subproject('sleef')
   sleef_dep = sleef_subproj.get_variable('sleef_dep')
   sleefquad_dep = sleef_subproj.get_variable('sleefquad_dep')
+  tlfloat_dep = sleef_subproj.get_variable('tlfloat_dep')
   warning(fallback_reason)
   message('Proceeding with vendored SLEEF subproject instead')
 endif
@@ -84,7 +89,7 @@ message('Using NumPy version: @0@'.format(numpy_version))
 npymath_path = incdir_numpy / '..' / 'lib'
 npymath_lib = c.find_library('npymath', dirs: npymath_path)
 
-dependencies = [py_dep, qblas_dep, sleef_dep, sleefquad_dep, npymath_lib]
+dependencies = [py_dep, qblas_dep, sleef_dep, sleefquad_dep, tlfloat_dep, npymath_lib]
 
 # Add OpenMP dependency (optional, for threading)
 openmp_dep = dependency('openmp', required: false, static: false)
@@ -99,6 +104,11 @@ if not is_windows
     if cpp.has_argument('-fext-numeric-literals')
         add_project_arguments('-fext-numeric-literals', language: 'cpp')
     endif
+
+    # Suppress warnings from system headers (sleefquad.h has many unused inline functions)
+    if cpp.has_argument('-Wno-unused-function')
+        add_project_arguments('-Wno-unused-function', language: ['c', 'cpp'])
+    endif
 endif
 
 # Thread-local storage detection (borrowed from NumPy)

quaddtype/numpy_quaddtype/src/dragon4.c

@@ -26,6 +26,11 @@ Modifications are specific to support the SLEEF_QUAD
 #include "scalar.h"
 
 
+// Undefine NPY_TLS if already defined (avoid redefinition warning)
+#ifdef NPY_TLS
+    #undef NPY_TLS
+#endif
+
 #ifdef __cplusplus
     #define NPY_TLS thread_local
 #elif defined(HAVE_THREAD_LOCAL)
@@ -2005,8 +2010,6 @@ PyObject *
 Dragon4_Positional(PyObject *obj, DigitMode digit_mode, CutoffMode cutoff_mode, int precision,
                    int min_digits, int sign, TrimMode trim, int pad_left, int pad_right)
 {
-    npy_double v;
-
     if (PyObject_TypeCheck(obj, &QuadPrecision_Type)) {
         QuadPrecisionObject *quad_obj = (QuadPrecisionObject *)obj;
         if (quad_obj->backend == BACKEND_SLEEF) {
@@ -2028,8 +2031,6 @@ PyObject *
 Dragon4_Scientific(PyObject *obj, DigitMode digit_mode, int precision, int min_digits, int sign,
                    TrimMode trim, int pad_left, int exp_digits)
 {
-    npy_double val;
-
     if (PyObject_TypeCheck(obj, &QuadPrecision_Type)) {
         QuadPrecisionObject *quad_obj = (QuadPrecisionObject *)obj;
         if (quad_obj->backend == BACKEND_SLEEF) {

quaddtype/numpy_quaddtype/src/ops.hpp

@@ -13,13 +13,13 @@ typedef Sleef_quad (*unary_op_quad_def)(const Sleef_quad *);
 // Unary Quad operations with 2 outputs (for modf, frexp)
 typedef void (*unary_op_2out_quad_def)(const Sleef_quad *, Sleef_quad *, Sleef_quad *);
 
-static Sleef_quad
+[[maybe_unused]] static Sleef_quad
 quad_negative(const Sleef_quad *op)
 {
     return Sleef_negq1(*op);
 }
 
-static Sleef_quad
+[[maybe_unused]] static Sleef_quad
 quad_positive(const Sleef_quad *op)
 {
     return *op;
@@ -929,15 +929,15 @@ static inline Sleef_quad quad_set_words64(int64_t hx, uint64_t lx)
 static inline Sleef_quad
 quad_nextafter(const Sleef_quad *x, const Sleef_quad *y)
 {
-    int64_t hx, hy, ix, iy;
+    int64_t hx, hy, ix;
     uint64_t lx, ly;
 
     quad_get_words64(&hx, &lx, *x);
     quad_get_words64(&hy, &ly, *y);
 
     // extracting absolute value
     ix = hx & 0x7fffffffffffffffLL;
-    iy = hy & 0x7fffffffffffffffLL;
+    (void)ly;  // unused but needed for quad_get_words64
 
     // NaN if either is NaN
     if (Sleef_iunordq1(*x, *y)) {

quaddtype/numpy_quaddtype/src/scalar.c

@@ -191,12 +191,21 @@ QuadPrecision_from_object(PyObject *value, QuadBackendType backend)
     else if (PyUnicode_Check(value)) {
         const char *s = PyUnicode_AsUTF8(value);
         char *endptr = NULL;
-        int err = cstring_to_quad(s, backend, &self->value, &endptr, true);
+        int err = NumPyOS_ascii_strtoq(s, backend, &self->value, &endptr);
         if (err < 0) {
             PyErr_SetString(PyExc_ValueError, "Unable to parse string to QuadPrecision");
             Py_DECREF(self);
             return NULL;
         }
+        // Skip trailing whitespace (matches Python's float() behavior)
+        while (ascii_isspace(*endptr)) {
+            endptr++;
+        }
+        if (*endptr != '\0') {
+            PyErr_SetString(PyExc_ValueError, "Unable to parse string to QuadPrecision");
+            Py_DECREF(self);
+            return NULL;
+        }
     }
     else if (PyBytes_Check(value)) {
         const char *s = PyBytes_AsString(value);
@@ -205,12 +214,21 @@ QuadPrecision_from_object(PyObject *value, QuadBackendType backend)
             return NULL;
         }
         char *endptr = NULL;
-        int err = cstring_to_quad(s, backend, &self->value, &endptr, true);
+        int err = NumPyOS_ascii_strtoq(s, backend, &self->value, &endptr);
         if (err < 0) {
             PyErr_SetString(PyExc_ValueError, "Unable to parse bytes to QuadPrecision");
             Py_DECREF(self);
             return NULL;
         }
+        // Skip trailing whitespace (matches Python's float() behavior)
+        while (ascii_isspace(*endptr)) {
+            endptr++;
+        }
+        if (*endptr != '\0') {
+            PyErr_SetString(PyExc_ValueError, "Unable to parse bytes to QuadPrecision");
+            Py_DECREF(self);
+            return NULL;
+        }
     }
     else if (PyLong_Check(value)) {
         return quad_from_py_int(value, backend, self);
@@ -320,19 +338,6 @@ QuadPrecision_str(QuadPrecisionObject *self)
     return PyUnicode_FromString(buffer);
 }
 
-static PyObject *
-QuadPrecision_repr(QuadPrecisionObject *self)
-{
-    PyObject *str = QuadPrecision_str(self);
-    if (str == NULL) {
-        return NULL;
-    }
-    const char *backend_str = (self->backend == BACKEND_SLEEF) ? "sleef" : "longdouble";
-    PyObject *res = PyUnicode_FromFormat("QuadPrecision('%S', backend='%s')", str, backend_str);
-    Py_DECREF(str);
-    return res;
-}
-
 static PyObject *
 QuadPrecision_repr_dragon4(QuadPrecisionObject *self)
 {

quaddtype/numpy_quaddtype/src/umath/umath.cpp

@@ -25,60 +25,6 @@ extern "C" {
 #include "comparison_ops.h"
 #include "matmul.h"
 
-// helper debugging function
-static const char *
-get_dtype_name(PyArray_DTypeMeta *dtype)
-{
-    if (dtype == &QuadPrecDType) {
-        return "QuadPrecDType";
-    }
-    else if (dtype == &PyArray_BoolDType) {
-        return "BoolDType";
-    }
-    else if (dtype == &PyArray_ByteDType) {
-        return "ByteDType";
-    }
-    else if (dtype == &PyArray_UByteDType) {
-        return "UByteDType";
-    }
-    else if (dtype == &PyArray_ShortDType) {
-        return "ShortDType";
-    }
-    else if (dtype == &PyArray_UShortDType) {
-        return "UShortDType";
-    }
-    else if (dtype == &PyArray_IntDType) {
-        return "IntDType";
-    }
-    else if (dtype == &PyArray_UIntDType) {
-        return "UIntDType";
-    }
-    else if (dtype == &PyArray_LongDType) {
-        return "LongDType";
-    }
-    else if (dtype == &PyArray_ULongDType) {
-        return "ULongDType";
-    }
-    else if (dtype == &PyArray_LongLongDType) {
-        return "LongLongDType";
-    }
-    else if (dtype == &PyArray_ULongLongDType) {
-        return "ULongLongDType";
-    }
-    else if (dtype == &PyArray_FloatDType) {
-        return "FloatDType";
-    }
-    else if (dtype == &PyArray_DoubleDType) {
-        return "DoubleDType";
-    }
-    else if (dtype == &PyArray_LongDoubleDType) {
-        return "LongDoubleDType";
-    }
-    else {
-        return "UnknownDType";
-    }
-}
-
 int
 init_quad_umath(void)
 {

quaddtype/numpy_quaddtype/src/utilities.c

@@ -134,12 +134,17 @@ NumPyOS_ascii_strtoq(const char *s, QuadBackendType backend, quad_value *out_val
             }
         }
 
-        // Set NaN value (sign is ignored for NaN)
+        // Set NaN value with sign preserved
         if (backend == BACKEND_SLEEF) {
-            out_value->sleef_value = QUAD_PRECISION_NAN;
+            Sleef_quad nan_val = QUAD_PRECISION_NAN;
+            // Apply sign to NaN (negative NaN has sign bit set)
+            if (sign < 0) {
+                nan_val = Sleef_negq1(nan_val);
+            }
+            out_value->sleef_value = nan_val;
         }
         else {
-            out_value->longdouble_value = nanl("");
+            out_value->longdouble_value = sign < 0 ? -nanl("") : nanl("");
         }
 
         if (endptr) {
@@ -157,15 +162,98 @@ int cstring_to_quad(const char *str, QuadBackendType backend, quad_value *out_va
 char **endptr, bool require_full_parse)
 {
   if(backend == BACKEND_SLEEF) {
-    out_value->sleef_value = Sleef_strtoq(str, endptr);
+    // SLEEF 4.0's Sleef_strtoq doesn't properly set endptr to indicate
+    // where parsing stopped. It always sets endptr to the end of the string.
+    // We need to manually validate and track the parse position.
+
+    const char *p = str;
+
+    // Skip leading whitespace
+    while (ascii_isspace(*p)) {
+      p++;
+    }
+
+    // Handle optional sign
+    if (*p == '+' || *p == '-') {
+      p++;
+    }
+
+    // Must have at least one digit or decimal point followed by digit
+    int has_digits = 0;
+
+    // Parse integer part
+    while (ascii_isdigit(*p)) {
+      has_digits = 1;
+      p++;
+    }
+
+    // Parse decimal point and fractional part
+    if (*p == '.') {
+      p++;
+      while (ascii_isdigit(*p)) {
+        has_digits = 1;
+        p++;
+      }
+    }
+
+    // Must have at least one digit somewhere
+    if (!has_digits) {
+      if (endptr) *endptr = (char *)str;
+      return -1;
+    }
+
+    // Parse optional exponent
+    if (*p == 'e' || *p == 'E') {
+      const char *exp_start = p;
+      p++;
+
+      // Optional sign in exponent
+      if (*p == '+' || *p == '-') {
+        p++;
+      }
+
+      // Must have at least one digit in exponent
+      if (!ascii_isdigit(*p)) {
+        // Invalid exponent, backtrack
+        p = exp_start;
+      } else {
+        while (ascii_isdigit(*p)) {
+          p++;
+        }
+      }
+    }
+
+    // Now p points to where valid parsing ends
+    // SLEEF 4.0's Sleef_strtoq has a bug where it doesn't properly stop at whitespace
+    // or other delimiters. We need to create a null-terminated substring.
+    size_t len = p - str;
+    char *temp = (char *)malloc(len + 1);
+    if (!temp) {
+      if (endptr) *endptr = (char *)str;
+      return -1;
+    }
+    memcpy(temp, str, len);
+    temp[len] = '\0';
+
+    // Call Sleef_strtoq with the bounded string
+    char *sleef_endptr;
+    out_value->sleef_value = Sleef_strtoq(temp, &sleef_endptr);
+    free(temp);
+
+    // Set endptr to our calculated position
+    if (endptr) {
+      *endptr = (char *)p;
+    }
+
   } else {
     out_value->longdouble_value = strtold(str, endptr);
   }
-  if(*endptr == str) 
+
+  if(endptr && *endptr == str) 
     return -1; // parse error - nothing was parsed
 
   // If full parse is required
-  if(require_full_parse && **endptr != '\0')
+  if(require_full_parse && endptr && **endptr != '\0')
     return -1; // parse error - characters remain to be converted
 
   return 0; // success

quaddtype/reinstall.sh

@@ -1,12 +1,11 @@
 #!/bin/bash
 set -x
 
-if [ -d "build/" ]; then
-    rm -r build
-    rm -rf dist/
-    rm -rf subprojects/qblas
-    rm -rf subprojects/sleef
-fi
+rm -rf build/
+rm -rf dist/
+rm -rf subprojects/qblas
+rm -rf subprojects/sleef
+rm -rf .mesonpy-*
 
 python -m pip uninstall -y numpy_quaddtype
 python -m pip install . -vv --no-build-isolation 2>&1 | tee build_log.txt