Re: [PATCH 1/3] ntdll/tests: Add tests for RtlIpv4StringToAddressEx, RtlIpv6StringToAddress (try 2)
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Hi,
I've already talked to Mark about this on #winehackers, just repeating it for the list: My remaining request with these tests is that I believe we should test the full set of inputs against the Ex and non-Ex versions and ideally reuse the test table (for base and ex, not for ipv4 and ipv6). That will tell us if there's any difference in the IP parsing part of the base and Ex functions. It'll also be interesting how the base function handles ports in the string, and, in case of ipv6, brackets around the IP.
Other than that two minor style issues: Change the subject of patch one (it only handles RtlIpv4StringToAddressEx), and you still have different levels of indentation in the line continuations of some ok() statements.
Cheers, Stefan
Am 2015-02-15 um 18:02 schrieb Mark Jansen:
Add test for RtlIpv4StringToAddressEx
--- dlls/ntdll/tests/rtl.c | 101 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 101 insertions(+)
Hello,
I do not agree with this (combining the test into one table): Any sensible implementation will have the Ex function call the non-Ex function to handle the ip parsing part, in which case it does not serve any purpose to run all tests again for the Ex function (essentially just re-testing the non-Ex function). Besides that, sharing the table between the functions means having separate results for the Ex and non-Ex in the same table, and what is the point of one shared table then?
regards,
On Mon, Feb 16, 2015 at 12:02 PM, Stefan Dösinger stefandoesinger@gmail.com wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Hi,
I've already talked to Mark about this on #winehackers, just repeating it for the list: My remaining request with these tests is that I believe we should test the full set of inputs against the Ex and non-Ex versions and ideally reuse the test table (for base and ex, not for ipv4 and ipv6). That will tell us if there's any difference in the IP parsing part of the base and Ex functions. It'll also be interesting how the base function handles ports in the string, and, in case of ipv6, brackets around the IP.
Other than that two minor style issues: Change the subject of patch one (it only handles RtlIpv4StringToAddressEx), and you still have different levels of indentation in the line continuations of some ok() statements.
Cheers, Stefan
Am 2015-02-15 um 18:02 schrieb Mark Jansen:
Add test for RtlIpv4StringToAddressEx
--- dlls/ntdll/tests/rtl.c | 101 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 101 insertions(+)
-----BEGIN PGP SIGNATURE----- Version: GnuPG v2
iQIcBAEBAgAGBQJU4c5HAAoJEN0/YqbEcdMwnPIP/RT2i0fG4cPqKikwa26VbRyn EG2yeG78Wjx/vysLuQfGe8alhUycGrgMzW4N4m8QiAx/xNFv4eX0OdI+5gYMbSHK dYZSzTSKbqoJbnBlZwt1MoaaNceX/zIEy2WoaRCYpsUJfpscT/sOdJ+/gW55gDkH 605vvruhgJFSFVeZDVzfR9gQcY51iHB0KDemX2p1mwHElr6+HwYSHOQt+8KCL8YV wfC87NSMEAfEjIudYPL7SFHvUuLr3GoHBBt57R9p/2gNT7eJ7ll9ANvShFkRENjL jmKJQj7exehqrAGNWBqbeIPSjHBQoGl/UZPwIE39UioNhRo+sWi5A8tBbUMyL3o5 VENuxL987kU1L9Xpa6/jEcrQksgafCa742/wWSAjPGE0D2iTUr0hQN0zQoioTPA/ zuaDyDO3CBCyFaF0A9h9FzxOkuRAHo6TSKw0IYOrI2XENAoqQ+B+fX0ll9t/5Tu0 iUIea06yp2/fRButo58sJNAd3dUZpRKf1qT+Eu7Cn+AJfgi3TsKY/Lw0gvyb4XJh qOdp7EPzqY1PURn6+Z8aDwNZRA2du3NM91SSMfwz+qwZtakf8K/XtkpC1CpNLONJ kfltxHVALBKT0NTDzvprrSzvTVH90hJLE+nX/E5Qf6SzeRVqxsPomBtOj2VLuapZ 4ehIhXsaEJJz9jc31tr9 =E0dA -----END PGP SIGNATURE-----
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Hi,
Am 2015-02-19 um 11:00 schrieb Mark Jansen:
I do not agree with this (combining the test into one table): Any sensible implementation will have the Ex function call the non-Ex function to handle the ip parsing part,
The way the terminator parameter works is a strong indicator of that, yeah. But the usual way is for the non-Ex function to call the Ex one. Microsoft also loves to duplicate code and functionality (see e.g. gdi vs gdiplus vs dwrite). I guess our implementation can be built that way, but I don't think the tests should be written based on that assumption. Instead, the tests should try their best to show that this is how the relation between base and ex works.
I guess the "terminator" output helps the caller (in your case the -Ex function) find out how many characters the base function read and where it has to look for the port. Though there are still some potential trouble cases like "192.168.0.1 :80"
in which case it does not serve any purpose to run all tests again for the Ex function (essentially just re-testing the non-Ex function).
Running the extra tests is not much of a problem. And the full set of tests needs to be written anyway.
Besides that, sharing the table between the functions means having separate results for the Ex and non-Ex in the same table, and what is the point of one shared table then?
If there is no reusable overlap then it obviously makes no sense. But my understanding is that for any input that does not contain a port, non-ex and -ex will behave in the same way. So I expect that you can share IP address based tests like 256.0.0.1 between both. You have to test how the non-ex version reacts to port information - these sets of input data are needed for both versions too. That will mostly be the success cases of the -Ex tests.
Of course there will be some tests (e.g. port corner cases that fail in -Ex) that only make sense for the Ex version. I don't see a problem with running them against the base version and checking that you indeed get STATUS_INVALID_PARAMETER back.
Fwiw, if you still think that merging the tables doesn't make sense I'll trust you on that - after all, you are writing the code. But please give it some serious thought.
-
Mark Jansen -
Stefan Dösinger