[thomasbruggink]

Fix the bug where the following:

struct C {
    1: i32 value
}

struct A {
    1: C nested = {}
}

struct B {
    1: A itm = {}
}

would generate the following invalid python code:

class C:
    thrift_spec = None

    def __init__(self, value=None):
        self.value = value

class A:
    thrift_spec = None

    def __init__(self, nested=C()):
        if nested is self.thrift_spec[1][4]:
            nested = C()
        self.nested = nested

class B:
    thrift_spec = None

    def __init__(self, itm=A()):
        if itm is self.thrift_spec[1][4]:
            itm = A()
        self.itm = itm

The problem with the above code is that self.thrift_spec was initialized and set after the module was loaded. In python this would work when B is instantiated but since B immediately invokes A() it happens before the thrift_spec is set and this would result in TypeError: 'NoneType' object is not subscriptable. By first initializing a temporary placeholder with a unique address to compare to called _THRIFT_DEFAULT the generated code would look like:

_THRIFT_DEFAULT = object()

class C:
    def __init__(self, value=None):
        self.value = value

class A:
    def __init__(self, nested=_THRIFT_DEFAULT):
        if nested is _THRIFT_DEFAULT:
            nested = C()

class B:
    def __init__(self, itm=_THRIFT_DEFAULT):
        if itm is _THRIFT_DEFAULT:
            itm = A()

Here the instantiation of the default object is defered until after the module has been fully loaded resulting in the correct value being assigned.

• Did you create an Apache Jira ticket? (Request account here, not required for trivial changes)
• If a ticket exists: Does your pull request title follow the pattern "THRIFT-NNNN: describe my issue"?
• Did you squash your changes to a single commit? (not required, but preferred)
• Did you do your best to avoid breaking changes? If one was needed, did you label the Jira ticket with "Breaking-Change"?
• If your change does not involve any code, include [skip ci] anywhere in the commit message to free up build resources.

[Jens-G]

Code review

No blocking issues found. Checked for bugs and CLAUDE.md compliance.

A few suggestions, below the bar for an issue but verified:

• A small formality for when you next push: CONTRIBUTING.md asks for a colon after the ticket id in the commit subject (THRIFT-4623: Fix python complex nested struct includes, the PR title can follow suit) and a Client: py line below it.

thrift/CONTRIBUTING.md

Lines 18 to 27 in bb4bf37

	1. All pull requests should contain a single commit per issue, or we will ask you to squash it.
	1. The pull request title must begin with the Jira THRIFT ticket identifier if it has an associated ticket, for example:
	THRIFT-9999: an example pull request title
	1. Commit messages must follow this pattern for code changes (deviations will not be merged):
	THRIFT-9999: [summary of fix, one line if possible]
	Client: [language(s) affected, comma separated, for example: "cpp,erl,perl"]

• type now goes through get_true_type(), and the same variable feeds the unchanged immutable branch below it. With -gen py:enum, a field typed as a typedef of an enum, in an exception or a python.immutable struct, now gets the enum coercion (x if hasattr(x, 'value') else Color.__members__.get(x)) instead of a plain assignment, so an int passed there is stored as None. Non-typedef enum fields already behave that way, so this may be acceptable, but it is a behaviour change outside the struct-default fix. A separate variable for the new check would avoid it.

thrift/compiler/cpp/src/thrift/generate/t_py_generator.cc

Lines 910 to 926 in bb4bf37

	// Initialize fields
	t_type* type = get_true_type((*m_iter)->get_type());
	if (!type->is_base_type() && !type->is_enum() && (*m_iter)->get_value() != nullptr) {
	indent(out) << "if " << maybe_escape_identifier((*m_iter)->get_name()) << " is _THRIFT_DEFAULT:" << '\n';
	indent_up();
	indent(out) << maybe_escape_identifier((*m_iter)->get_name()) << " = " << render_field_default_value(*m_iter)
	<< '\n';
	indent_down();
	}
	if (is_immutable(tstruct)) {
	if (gen_enum_ && type->is_enum()) {
	indent(out) << "super(" << maybe_escape_identifier(tstruct->get_name()) << ", self).__setattr__('"
	<< (*m_iter)->get_name() << "', " << maybe_escape_identifier((*m_iter)->get_name())
	<< " if hasattr(" << maybe_escape_identifier((*m_iter)->get_name()) << ", 'value') else "
	<< type_name(type) << ".__members__.get(" << maybe_escape_identifier((*m_iter)->get_name()) << "))" << '\n';
	} else if (gen_newstyle_ || gen_dynamic_) {

• With -gen py:type_hints, the new default produces signatures like nested: typing.Optional[C] = _THRIFT_DEFAULT. A type checker rejects that, because object() is neither C nor None. Runtime behaviour is not affected.

thrift/compiler/cpp/src/thrift/generate/t_py_generator.cc

Lines 2786 to 2794 in bb4bf37

	result << " = ";
	if (tfield->get_value() != nullptr) {
	t_type* type = get_true_type(tfield->get_type());
	if (!type->is_base_type() && !type->is_enum()) {
	result << "_THRIFT_DEFAULT";
	} else {
	result << render_field_default_value(tfield);
	}
	} else {

🤖 Generated with Claude Code

[Jens-G]

@thomasbruggink Coulkd you please have a look? Thanks!

[thomasbruggink]

@thomasbruggink Coulkd you please have a look? Thanks!

@Jens-G i can take a look at the end of next week, currently on vacation without access to a computer 🙇 thanks for the review

[thomasbruggink]

@Jens-G updated, sorry for the delay 🙇

[Jens-G]

Code review

Re-reviewed at 5513cbaf6. No blocking issues found. Checked for bugs and CLAUDE.md compliance.

The three points from the last round are addressed.

One suggestion, below the bar for an issue but verified:

• A struct or container default is now built inside __init__, where every parameter shadows a module-level name. If a field has the same name as the module its default refers to, construction fails. With include "shared.thrift" and struct Wrapper { 1: shared.SharedStruct shared = {} }, this head generates

  def __init__(self, shared = _THRIFT_DEFAULT,):
      if shared is _THRIFT_DEFAULT:
          shared = shared.ttypes.SharedStruct(**{
          })

  and Wrapper() raises AttributeError: 'object' object has no attribute 'ttypes'. Reading a message that contains a Wrapper fails the same way, because read() constructs it first. include "address.thrift" with 1: address.Address address = {} behaves the same. I built compilers from this head and from the merge base: the merge base's output constructs both structs and round-trips a struct that contains a Wrapper, this head's output fails all three. The parameter names are the keyword arguments, so the default needs a name that a field cannot shadow, for example a module-level factory per field.

thrift/compiler/cpp/src/thrift/generate/t_py_generator.cc

Lines 914 to 920 in 5513cba

	t_type* true_type = get_true_type(type);
	if (!true_type->is_base_type() && !true_type->is_enum() && (*m_iter)->get_value() != nullptr) {
	indent(out) << "if " << maybe_escape_identifier((*m_iter)->get_name()) << " is _THRIFT_DEFAULT:" << '\n';
	indent_up();
	indent(out) << maybe_escape_identifier((*m_iter)->get_name()) << " = " << render_field_default_value(*m_iter)
	<< '\n';
	indent_down();

🤖 Generated with Claude Code