On Sun, Jan 21, 2018 at 12:45 AM, Nikolay Sivov <nsivov@codeweavers.com> wrote:

On 01/20/2018 08:40 AM, Zhiyi Zhang wrote:

Hi Nikolay,

This fixes https://bugs.winehq.org/show_bug.cgi?id=35127

As I was working through bug 35127, I noticed that you had once changed the layout of struct _TREEITEM in commit 6ec621e835e03f715d5fee27c5de5b0d361814de.

Now I am sure that textWidth in struct _TREEITEM is a WORD and is at the
offset 0x1a. You can check out the bug comments for details. I tried to find out what is at 0x18 but no luck, thus I am adding a unknown member to make textWidth at the right offset.

������What do you think of the patch? Do you know what member is at struct _TREEITEM+0x18? I can use your opinion with this patch.



struct������_ITEM_DATA
{
-������ HTREEITEM parent; /* for root value of parent field is unidetified */
+������ HTREEITEM parent; /* for root value of parent field is unidentified */
������ HTREEITEM nextsibling;
������ HTREEITEM firstchild;
+������ UINT unknownField0; /* untested */
+������ UINT unknownField1; /* untested */
+������ UINT unknownField2; /* untested */
+������ WORD padding0; /* unknown field to make textWidth offset compatible with the native one */
+������ WORD textWidth; /* horizontal text extent for pszText */
};

They are all unknown, so a single unknown[7] is better, assuming this will work on 64 bits.

+������ /* padding0 should be unused on Wine */
+������ if(!strcmp(winetest_platform, "wine"))
+������ {
+������ ok_(__FILE__, line)(data->padding0 == 0, "padding0 %04x, got %04x\n", 0, data->padding0);
+������ ok_(__FILE__, line)(data->textWidth == textWidth, "textWidth %04x, got %04x\n", textWidth, data->textWidth);
+������ }
+������ /* TODO: textWidth computation on Windows is alway that of Wine plus 4, remove this test when it got corrected.
+������ Also, in Windows textWidth is calculated even if tree item is invisible whereas in Wine is postponed */
+������ else
+������ {
+������ ok_(__FILE__, line)(textWidth == 0 || data->textWidth == 0 || data->textWidth == textWidth + 4,
+������ "textWidth %04x, got %04x\n", textWidth, data->textWidth);
+������ }

This is very dirty. You should get rid of platform check; testing padding0 does not look useful; and last test line looks suspicious, if expected width is 0 nothing gets tested.

+������ /* Set font to make textWidth calculation consistent */
+������ hdc=GetDC(hMainWnd);
+������ SystemParametersInfoA(SPI_GETICONTITLELOGFONT, sizeof(lf), &lf, 0);
+������ SelectObject(hdc, CreateFontIndirectA(&lf));

This doesn't make it consistent, because you select one font, and control could be using another one, also you're using parent window for some reason. WM_GETFONT returns what you need.

Instead of calculating this width manually, it's better to use TVM_GETITEMRECT (with wparam == 1), if that gives expected width. If text portion rectangle width returned by this message matches internal field value on Windows, then we should use that. This will also hide +4 difference you see, unless it was caused by a wrong font.

typedef������struct _TREEITEM /* HTREEITEM is a _TREEINFO *. */
������ UINT callbackMask;
������ UINT state;
������ UINT stateMask;
+������ WORD padding0; /* unknown field to make textWidth offset compatible with the native one */
+������ WORD textWidth; /* horizontal text extent for pszText */

Instead this should match structure used in tests, i.e. it should start with same fields and not have incompatible data mixed with what we try to match.