Created attachment 6506 [details]
Standalone testcase.

I found a strange behaviour trying to implement a pretty printer. Attached is a test case (just unpack and run make). The issue is reproducible both with 7.4.1 release and with CVS head as of now (7.4.50.20120701-cvs). (It's also present in Debian testing and Ubuntu 12.04.)

Assumming a C++ class:

struct Foo
{
	int *m_ptr;
};

the following pretty printing class:

class MyTestPrinter:
	def __init__(self, val):
		self.val = val

	def get_ptr(self):
		return self.val['m_ptr']

	def children(self):
		print "foo"

	def to_string(self):
		return "0x%08x" % self.get_ptr()

causes a strange error message:

$1 = 0x7fffffffe390Traceback (most recent call last):
  File "test-printer.py", line 11, in children
    print "foo"
  File "<string>", line 13, in write
gdb.error: Cannot convert value to int.


The body of MyTestPrinter.to_string() might be wrong (using a gdb.Value for a pointer with %08x), but the actual error message complains about the 'print "foo"'. But the print is ok, so the error message is highly confusing.

To make this matter more mysterious, doing any of the following makes the error disappear:

* remove MyTestPrinter.children()
* add a long cast to MyTestPrinter.to_string(): return "0x%08x" % long(self.get_ptr())

So, if MyTestPrinter.to_string() is wrong, it should produce an appropriate error message, and it should be consistent, no matter if there's any MyTestPrinter.children() or not. Or, if MyTestPrinter.to_string() is correct, then the print in MyTestPrinter.children() should not cause any error.

It seems that it's not caused by print (in MyTestPrinter.children()) itself, but by the string literal, as I can get the same kind of error message for other things involving string literals in MyTestPrinter.children().