golang, errors, fmt: Error chaining (C++/Pyx)

Following errors model in Go, let's add support for errors to wrap other
errors and to be inspected/unwrapped:

- an error can additionally provide way to unwrap itself, if it
  implements errorWrapper interface;
- errors.Unwrap(err) tries to extract wrapped error;
- errors.Is(err) tests whether an item in error's chain matches target;
- `fmt.errorf("... : %w", ... err)` is similar to `fmt.errorf("... : %s", ... err.c_str())`
  but resulting error, when unwrapped, will return err.

Add C++ implementation for the above + tests.
Python analogs will follow in the next patches.

Top-level documentation is TODO.

See https://blog.golang.org/go1.13-errors for error chaining overview.

golang/.gitignore

 /_context.cpp
 /_cxx_test.cpp
+/_errors.cpp
 /_errors_test.cpp
+/_fmt.cpp
 /_fmt_test.cpp
 /_golang.cpp
 /_golang_test.cpp

golang/_errors.pxd

 # cython: language_level=2
-# Copyright (C) 2019  Nexedi SA and Contributors.
+# Copyright (C) 2019-2020  Nexedi SA and Contributors.
 #                          Kirill Smelkov <kirr@nexedi.com>
 #
 # This program is free software: you can Use, Study, Modify and Redistribute
@@ -20,11 +20,16 @@
 """Package errors mirrors Go package errors.
 
  - `New` creates new error with provided text.
+ - `Unwrap` tries to extract wrapped error.
+ - `Is` tests whether an item in error's chain matches target.
 
 See also https://golang.org/pkg/errors for Go errors package documentation.
+See also https://blog.golang.org/go1.13-errors for error chaining overview.
 """
 
-from golang cimport error, string
+from golang cimport error, string, cbool
 
 cdef extern from "golang/errors.h" namespace "golang::errors" nogil:
     error New(const string& text)
+    error Unwrap(error err)
+    cbool Is(error err, error target)

golang/_errors_test.pyx

@@ -30,9 +30,19 @@ from golang cimport topyexc
 cdef extern from * nogil:
     """
     extern void _test_errors_new_cpp();
+    extern void _test_errors_unwrap_cpp();
+    extern void _test_errors_is_cpp();
     """
     void _test_errors_new_cpp()                 except +topyexc
+    void _test_errors_unwrap_cpp()              except +topyexc
+    void _test_errors_is_cpp()                  except +topyexc
 
 def test_errors_new_cpp():
     with nogil:
         _test_errors_new_cpp()
+def test_errors_unwrap_cpp():
+    with nogil:
+        _test_errors_unwrap_cpp()
+def test_errors_is_cpp():
+    with nogil:
+        _test_errors_is_cpp()

golang/_fmt.pxd

 # cython: language_level=2
-# Copyright (C) 2019  Nexedi SA and Contributors.
+# Copyright (C) 2019-2020  Nexedi SA and Contributors.
 #                          Kirill Smelkov <kirr@nexedi.com>
 #
 # This program is free software: you can Use, Study, Modify and Redistribute
@@ -22,9 +22,12 @@
  - `sprintf` formats text into string.
  - `errorf`  formats text into error.
 
-NOTE: formatting rules are those of libc, not Go.
+NOTE: with exception of %w, formatting rules are those of libc, not Go(*).
 
 See also https://golang.org/pkg/fmt for Go fmt package documentation.
+
+(*) errorf additionally handles Go-like %w to wrap an error similarly to
+    https://blog.golang.org/go1.13-errors .
 """
 
 from golang cimport string, error

golang/_golang.pxd

@@ -173,6 +173,15 @@ cdef extern from "golang/libgolang.h" namespace "golang" nogil:
         string Error    "_ptr()->Error" ()
 
 
+    # error wrapper interface
+    cppclass _errorWrapper (_error):
+        error Unwrap()
+
+    cppclass errorWrapper (refptr[_errorWrapper]):
+        # errorWrapper.X = errorWrapper->X in C++
+        error Unwrap    "_ptr()->Unwrap" ()
+
+
 # ---- python bits ----
 
 cdef void topyexc() except *

golang/_testing.h

@@ -35,6 +35,9 @@ using namespace golang;
 // string -> string (not in STL, huh ?!)
 string to_string(const string& s) { return s; }
 
+// error -> string
+string to_string(error err) { return (err == nil) ? "nil" : err->Error(); }
+
 // vector<T> -> string
 template<typename T>
 string to_string(const vector<T>& v) {

golang/errors.cpp

-// Copyright (C) 2019  Nexedi SA and Contributors.
+// Copyright (C) 2019-2020  Nexedi SA and Contributors.
 //                          Kirill Smelkov <kirr@nexedi.com>
 //
 // This program is free software: you can Use, Study, Modify and Redistribute
@@ -51,4 +51,32 @@ error New(const string& text) {
     return adoptref(static_cast<_error*>(new _TextError(text)));
 }
 
+
+error Unwrap(error err) {
+    if (err == nil)
+        return nil;
+
+    _errorWrapper* _werr = dynamic_cast<_errorWrapper*>(err._ptr());
+    if (_werr == nil)
+        return nil;
+
+    return _werr->Unwrap();
+}
+
+bool Is(error err, error target) {
+    if (target == nil)
+        return (err == nil);
+
+    for(;;) {
+        if (err == nil)
+            return false;
+
+        if (typeid(*err) == typeid(*target))
+            if (err->Error() == target->Error()) // XXX hack instead of dynamic == (not available in C++)
+                return true;
+
+        err = Unwrap(err);
+    }
+}
+
 }}  // golang::errors::

golang/errors.h

 #ifndef _NXD_LIBGOLANG_ERRORS_H
 #define _NXD_LIBGOLANG_ERRORS_H
 
-// Copyright (C) 2019  Nexedi SA and Contributors.
+// Copyright (C) 2019-2020  Nexedi SA and Contributors.
 //                          Kirill Smelkov <kirr@nexedi.com>
 //
 // This program is free software: you can Use, Study, Modify and Redistribute
@@ -23,8 +23,11 @@
 // Package errors mirrors Go package errors.
 //
 //  - `New` creates new error with provided text.
+//  - `Unwrap` tries to extract wrapped error.
+//  - `Is` tests whether an item in error's chain matches target.
 //
 // See also https://golang.org/pkg/errors for Go errors package documentation.
+// See also https://blog.golang.org/go1.13-errors for error chaining overview.
 
 #include <golang/libgolang.h>
 
@@ -35,6 +38,15 @@ namespace errors {
 // New creates new error with provided text.
 LIBGOLANG_API error New(const string& text);
 
+// Unwrap tries to unwrap error.
+//
+// If err implements Unwrap method, it returns err.Unwrap().
+// Otherwise it returns nil.
+LIBGOLANG_API error Unwrap(error err);
+
+// Is returns whether target matches any error in err's error chain.
+LIBGOLANG_API bool Is(error err, error target);
+
 }}  // golang::errors::
 
 #endif  // _NXD_LIBGOLANG_ERRORS_H

golang/errors_test.cpp

@@ -26,3 +26,84 @@ void _test_errors_new_cpp() {
     error err = errors::New("hello world");
     ASSERT_EQ(err->Error(), "hello world");
 }
+
+struct _MyError final : _error, object {
+    string msg;
+    string Error() {
+        return msg;
+    }
+    _MyError(const string& msg) : msg(msg) {}
+    ~_MyError() {}
+
+    void incref() { object::incref(); }
+    void decref() {
+        if (object::__decref())
+            delete this;
+    }
+};
+
+struct _MyWrapError final : _errorWrapper, object {
+    string subj;
+    error  err;
+    string Error() {
+        return subj + ": " + err->Error();
+    }
+    error Unwrap() {
+        return err;
+    }
+
+    _MyWrapError(string subj, error err) : subj(subj), err(err) {}
+    ~_MyWrapError() {}
+
+    void incref() { object::incref(); }
+    void decref() {
+        if (object::__decref())
+            delete this;
+    }
+};
+
+typedef refptr<_MyError> MyError;
+
+void _test_errors_unwrap_cpp() {
+    error err1, err2;
+    ASSERT_EQ(errors::Unwrap(/*nil*/error()), /*nil*/error());
+
+    err1 = adoptref(static_cast<_error*>(new _MyError("zzz")));
+    ASSERT_EQ(errors::Unwrap(err1), /*nil*/error());
+
+    _MyWrapError* _err2 = new _MyWrapError("aaa", err1);
+    err2 = adoptref(static_cast<_error*>(_err2));
+    ASSERT_EQ(errors::Unwrap(err2), err1);
+
+    // test err2.Unwrap() returning nil
+    _err2->err = nil;
+    ASSERT_EQ(_err2->Unwrap(), /*nil*/error());
+    ASSERT_EQ(errors::Unwrap(err2), /*nil*/error());
+}
+
+void _test_errors_is_cpp() {
+    auto E = errors::New;
+    ASSERT_EQ(errors::Is(/*nil*/error(),    /*nil*/error()),    true);
+    ASSERT_EQ(errors::Is(E("a"),            /*nil*/error()),    false);
+    ASSERT_EQ(errors::Is(/*nil*/error(),    E("b")),            false);
+
+    auto W = [](string subj, error err) -> error {
+        return adoptref(static_cast<_error*>(new _MyWrapError(subj, err)));
+    };
+
+    error ewrap = W("hello", W("world", E("мир")));
+    ASSERT_EQ(errors::Is(ewrap, E("мир")),    true);
+    ASSERT_EQ(errors::Is(ewrap, E("май")),    false);
+
+    ASSERT_EQ(errors::Is(ewrap, W("world", E("мир"))),  true);
+    ASSERT_EQ(errors::Is(ewrap, W("hello", E("мир"))),  false);
+    ASSERT_EQ(errors::Is(ewrap, W("hello", E("май"))),  false);
+    ASSERT_EQ(errors::Is(ewrap, W("world", E("май"))),  false);
+
+    ASSERT_EQ(errors::Is(ewrap, W("hello", W("world", E("мир")))),  true);
+    ASSERT_EQ(errors::Is(ewrap, W("a",     W("world", E("мир")))),  false);
+    ASSERT_EQ(errors::Is(ewrap, W("hello", W("b",     E("мир")))),  false);
+    ASSERT_EQ(errors::Is(ewrap, W("hello", W("world", E("c")))),    false);
+
+    ASSERT_EQ(errors::Is(ewrap, W("x", W("hello", W("world", E("мир"))))),  false);
+}

golang/fmt.cpp

@@ -24,9 +24,11 @@
 
 #include "golang/fmt.h"
 #include "golang/errors.h"
+#include "golang/strings.h"
 
 #include <stdarg.h>
 #include <stdio.h>
+#include <string.h>
 
 
 // golang::fmt::
@@ -62,7 +64,7 @@ string sprintf(const char *format, ...) {
     return str;
 }
 
-error errorf(const string &format, ...) {
+error ___errorf(const string &format, ...) {
     va_list argp;
     va_start(argp, format);
     error err = errors::New(fmt::_vsprintf(format.c_str(), argp));
@@ -70,7 +72,7 @@ error errorf(const string &format, ...) {
     return err;
 }
 
-error errorf(const char *format, ...) {
+error ___errorf(const char *format, ...) {
     va_list argp;
     va_start(argp, format);
     error err = errors::New(fmt::_vsprintf(format, argp));
@@ -78,4 +80,90 @@ error errorf(const char *format, ...) {
     return err;
 }
 
+
+// _WrapError is the error created by errorf("...: %w", err).
+struct _WrapError final : _errorWrapper, object {
+    string _prefix;
+    error  _errSuffix;
+
+    _WrapError(const string& prefix, error err) : _prefix(prefix), _errSuffix(err) {}
+    error  Unwrap() { return _errSuffix; }
+    string Error() {
+        return _prefix + ": " +
+               (_errSuffix != nil ? _errSuffix->Error() : "%!w(<nil>)");
+    }
+
+    void incref() {
+        object::incref();
+    }
+    void decref() {
+        if (__decref())
+            delete this;
+    }
+    ~_WrapError() {}
+};
+
+// ___errorfTryWrap serves __errorf(format, last_err, ...headv)
+//
+// NOTE it is called with ... = original argv with last err converted to
+// err->Error().c_str() so that `errorf("... %s", ..., err)` also works.
+error ___errorfTryWrap(const string& format, error last_err, ...) {
+    error err;
+    va_list argp;
+    va_start(argp, last_err);
+
+    if (strings::has_suffix(format, ": %w")) {
+        err = adoptref(static_cast<_error*>(
+                    new _WrapError(
+                        fmt::_vsprintf(strings::trim_suffix(format, ": %w").c_str(), argp),
+                        last_err)));
+    }
+    else {
+        err = errors::New(fmt::_vsprintf(format.c_str(), argp));
+    }
+
+    va_end(argp);
+    return err;
+}
+
+error ___errorfTryWrap(const char *format, error last_err, ...) {
+    error err;
+    va_list argp;
+    va_start(argp, last_err);
+
+    const char *colon = strrchr(format, ':');
+    if (colon != NULL && strcmp(colon, ": %w") == 0) {
+        err = adoptref(static_cast<_error*>(
+                    new _WrapError(
+                        // TODO try to avoid std::string
+                        fmt::_vsprintf(strings::trim_suffix(format, ": %w").c_str(), argp),
+                        last_err)));
+    }
+    else {
+        err = errors::New(fmt::_vsprintf(format, argp));
+    }
+
+    va_end(argp);
+    return err;
+}
+
+
+// ___error_str is used by errorf to convert last_err into string for not %w.
+//
+// if we do not take nil into account the code crashes on nil->Error() call,
+// which is unreasonable, because both Go and C printf family print something
+// for nil instead of crash.
+//
+// string - not `const char*` - is returned, because returning
+// err->Error().c_str() would be not correct as error string might be destructed
+// at function exit, while if we return string, that would be correct and the
+// caller can use returned data without crashing.
+//
+// we don't return %!s(<nil>) since we don't know whether %s was used in format
+// tail or not.
+string ___error_str(error err) {
+    return (err != nil ? err->Error() : "(<nil>)");
+}
+
+
 }}  // golang::fmt::

golang/fmt.h

 #ifndef _NXD_LIBGOLANG_FMT_H
 #define _NXD_LIBGOLANG_FMT_H
 
-// Copyright (C) 2019  Nexedi SA and Contributors.
+// Copyright (C) 2019-2020  Nexedi SA and Contributors.
 //                          Kirill Smelkov <kirr@nexedi.com>
 //
 // This program is free software: you can Use, Study, Modify and Redistribute
@@ -25,11 +25,15 @@
 //  - `sprintf` formats text into string.
 //  - `errorf`  formats text into error.
 //
-// NOTE: formatting rules are those of libc, not Go.
+// NOTE: with exception of %w, formatting rules are those of libc, not Go(*).
 //
 // See also https://golang.org/pkg/fmt for Go fmt package documentation.
+//
+// (*) errorf additionally handles Go-like %w to wrap an error similarly to
+//     https://blog.golang.org/go1.13-errors .
 
 #include <golang/libgolang.h>
+#include <type_traits>
 
 // golang::fmt::
 namespace golang {
@@ -38,15 +42,108 @@ namespace fmt {
 // sprintf formats text into string.
 LIBGOLANG_API string sprintf(const string &format, ...);
 
+
+// intseq<i1, i2, ...> and intrange<n> are used by errorf to handle %w.
+namespace {
+    // intseq<i1, i2, ...> provides compile-time integer sequence.
+    // (std::integer_sequence is C++14 while libgolang targets C++11)
+    template<int ...nv>
+    struct intseq {
+        // apppend<x> defines intseq<i1, i2, ..., x>.
+        template<int x>
+        struct append {
+            using type = intseq<nv..., x>;
+        };
+    };
+
+    // intrange<n> provides integer sequence intseq<0, 1, 2, ..., n-1>.
+    template<int n>
+    struct intrange;
+
+    template<>
+    struct intrange<0> {
+        using type = intseq<>;
+    };
+
+    template<int n>
+    struct intrange {
+        using type = typename intrange<n-1>::type::template append<n-1>::type;
+    };
+}
+
 // `errorf`  formats text into error.
-LIBGOLANG_API error  errorf (const string &format, ...);
+//
+// format suffix ": %w" is additionally handled as in Go with
+// `errorf("... : %w", ..., err)` creating error that can be unwrapped back to err.
+LIBGOLANG_API error ___errorf(const string& format, ...);
+LIBGOLANG_API error ___errorfTryWrap(const string& format, error last_err, ...);
+LIBGOLANG_API string ___error_str(error err);
+
+// _errorf(..., err) tails here.
+template<typename ...Headv>
+inline error __errorf(std::true_type, const string& format, error last_err, Headv... headv) {
+    return ___errorfTryWrap(format, last_err, headv..., ___error_str(last_err).c_str());
+}
+
+// _errorf(..., !err) tails here.
+template<typename ...Headv, typename Last>
+inline error __errorf(std::false_type, const string& format, Last last, Headv... headv) {
+    return ___errorf(format, headv..., last);
+}
+
+template<typename ...Argv, int ...HeadIdxv>
+inline error _errorf(intseq<HeadIdxv...>, const string& format, Argv... argv) {
+    auto argt = std::make_tuple(argv...);
+    auto last = std::get<sizeof...(argv)-1>(argt);
+    return __errorf(std::is_same<decltype(last), error>(), format, last, std::get<HeadIdxv>(argt)...);
+}
+
+inline error errorf(const string& format) {
+    return ___errorf(format);
+}
+template<typename ...Argv>
+inline error errorf(const string& format, Argv... argv) {
+    return _errorf(typename intrange<sizeof...(argv)-1>::type(), format, argv...);
+}
+
+
 
 // `const char *` overloads just to catch format mistakes as
 // __attribute__(format) does not work with std::string.
 LIBGOLANG_API string sprintf(const char *format, ...)
                                 __attribute__ ((format (printf, 1, 2)));
-LIBGOLANG_API error  errorf (const char *format, ...)
-                                __attribute__ ((format (printf, 1, 2)));
+
+// cannot use __attribute__(format) for errorf as we add %w handling.
+// still `const char *` overload is useful for performance.
+LIBGOLANG_API error ___errorf(const char *format, ...);
+LIBGOLANG_API error ___errorfTryWrap(const char *format, error last_err, ...);
+
+// _errorf(..., err) tails here.
+template<typename ...Headv>
+inline error __errorf(std::true_type, const char *format, error last_err, Headv... headv) {
+    return ___errorfTryWrap(format, last_err, headv..., ___error_str(last_err).c_str());
+}
+
+// _errorf(..., !err) tails here.
+template<typename ...Headv, typename Last>
+inline error __errorf(std::false_type, const char *format, Last last, Headv... headv) {
+    return ___errorf(format, headv..., last);
+}
+
+template<typename ...Argv, int ...HeadIdxv>
+inline error _errorf(intseq<HeadIdxv...>, const char *format, Argv... argv) {
+    auto argt = std::make_tuple(argv...);
+    auto last = std::get<sizeof...(argv)-1>(argt);
+    return __errorf(std::is_same<decltype(last), error>(), format, last, std::get<HeadIdxv>(argt)...);
+}
+
+inline error errorf(const char *format) {
+    return ___errorf(format);
+}
+template<typename ...Argv>
+inline error errorf(const char *format, Argv... argv) {
+    return _errorf(typename intrange<sizeof...(argv)-1>::type(), format, argv...);
+}
 
 }}  // golang::fmt::
 

golang/fmt_test.cpp

@@ -18,31 +18,80 @@
 // See https://www.nexedi.com/licensing for rationale and options.
 
 #include "golang/fmt.h"
+#include "golang/errors.h"
 #include "golang/_testing.h"
 using namespace golang;
 
 void _test_fmt_sprintf_cpp() {
     // NOTE not using vargs helper, since sprintf itself uses vargs and we want
     // to test varg logic there for correctness too.
+    ASSERT_EQ(fmt::sprintf("")                   , "");
     ASSERT_EQ(fmt::sprintf("hello world")        , "hello world");
     ASSERT_EQ(fmt::sprintf("hello %d zzz", 123)  , "hello 123 zzz");
     ASSERT_EQ(fmt::sprintf("%s %s: %s", "read", "myfile", "myerror") , "read myfile: myerror");
 
     // with string format (not `const char *`)
-    const char *myerror = "myerror";
-    string f = "%s %s: %s";
+    ASSERT_EQ(fmt::sprintf(string(""))           , "");
     const char *myfile  = "myfile";
-    ASSERT_EQ(fmt::sprintf(f, "read", myfile, myerror) , "read myfile: myerror");
+    const char *myerror = "myerror";
+    ASSERT_EQ(fmt::sprintf(string("%s %s: %s"), "read", myfile, myerror) , "read myfile: myerror");
 }
 
 void _test_fmt_errorf_cpp() {
+    error myerr = errors::New("myerror");
+    error myerrWrap, myerr2;
+
+    ASSERT_EQ(fmt::errorf("")->Error()                   , "");
     ASSERT_EQ(fmt::errorf("hello world")->Error()        , "hello world");
     ASSERT_EQ(fmt::errorf("hello %d zzz", 123)->Error()  , "hello 123 zzz");
     ASSERT_EQ(fmt::errorf("%s %s: %s", "read", "myfile", "myerror")->Error() , "read myfile: myerror");
 
+    myerrWrap = fmt::errorf("%s %s: %w", "read", "myfile", myerr);
+    ASSERT_EQ(myerrWrap->Error(), "read myfile: myerror");
+    myerr2 = errors::Unwrap(myerrWrap);
+    ASSERT_EQ(myerr2, myerr);
+
+    myerrWrap = fmt::errorf("%s %s: %s", "read", "myfile", myerr);
+    ASSERT_EQ(myerrWrap->Error(), "read myfile: myerror");
+    myerr2 = errors::Unwrap(myerrWrap);
+    ASSERT_EQ(myerr2, /*nil*/error());
+
     // with string format (not `const char *`)
-    const char *myerror = "myerror";
-    string f = "%s %s: %s";
+    ASSERT_EQ(fmt::errorf(string(""))->Error()           , "");
+    const char *esmth = "esmth";
     const char *myfile = "myfile";
-    ASSERT_EQ(fmt::errorf(f, "read", myfile, myerror)->Error() , "read myfile: myerror");
+    ASSERT_EQ(fmt::errorf(string("%s %s: %s"), "read", myfile, esmth)->Error() , "read myfile: esmth");
+
+    myerrWrap = fmt::errorf(string("%s %s: %w"), "read", myfile, myerr);
+    ASSERT_EQ(myerrWrap->Error(), "read myfile: myerror");
+    myerr2 = errors::Unwrap(myerrWrap);
+    ASSERT_EQ(myerr2, myerr);
+
+    myerrWrap = fmt::errorf(string("%s %s: %s"), "read", myfile, myerr);
+    ASSERT_EQ(myerrWrap->Error(), "read myfile: myerror");
+    myerr2 = errors::Unwrap(myerrWrap);
+    ASSERT_EQ(myerr2, /*nil*/error());
+
+    // %w with nil
+    myerr = nil;
+    myerrWrap = fmt::errorf("%s %s: %w", "read", "myfile", myerr);
+    ASSERT_EQ(myerrWrap->Error(), "read myfile: %!w(<nil>)");
+    myerr2 = errors::Unwrap(myerrWrap);
+    ASSERT_EQ(myerr2, /*nil*/error());
+
+    myerrWrap = fmt::errorf(string("%s %s: %w"), "read", "myfile", myerr);
+    ASSERT_EQ(myerrWrap->Error(), "read myfile: %!w(<nil>)");
+    myerr2 = errors::Unwrap(myerrWrap);
+    ASSERT_EQ(myerr2, /*nil*/error());
+
+    // %s with nil
+    myerrWrap = fmt::errorf("%s %s: %s", "read", "myfile", myerr);
+    ASSERT_EQ(myerrWrap->Error(), "read myfile: (<nil>)");
+    myerr2 = errors::Unwrap(myerrWrap);
+    ASSERT_EQ(myerr2, /*nil*/error());
+
+    myerrWrap = fmt::errorf(string("%s %s: %s"), "read", "myfile", myerr);
+    ASSERT_EQ(myerrWrap->Error(), "read myfile: (<nil>)");
+    myerr2 = errors::Unwrap(myerrWrap);
+    ASSERT_EQ(myerr2, /*nil*/error());
 }

golang/libgolang.h

@@ -798,6 +798,12 @@ struct _error : _interface {
 };
 typedef refptr<_error> error;
 
+// an error can additionally provide Unwrap method if it wraps another error.
+struct _errorWrapper : _error {
+    virtual error Unwrap() = 0;
+};
+typedef refptr<_errorWrapper> errorWrapper;
+
 }   // golang::