GDI Line Drawing Quirk in NT 3.1 and 3.5
A program was looking somewhat odd in NT 3.1 and 3.5. Upon closer inspection, it turned out that the lines were slightly off.
According to the documentation of the Win32 LineTo() function, the line is drawn from the starting point inclusive to the ending point exclusive. This is true on Windows NT 3.51+, Windows 9x and even Win32s on Windows 3.1. On NT 3.1 and 3.5, however, the ending point is included. This can make GDI-drawn shapes look odd, as the lines may overshoot their intended connection point, especially for series of connected lines in different colours.
To test this, I wrote a Win32 program drawing a horizontal and vertical line. The horizontal line goes from point (11, 110) to (111 110) in the client area. The vertical line goes from (110, 11) to (110, 111) in the client area. Because GDI is not supposed to include the end point, and as only one coordinate changes per-pixel for each line, the last pixel of each line should be located at (110, 110), resulting in a perfect corner. Indeed, this was the case on Windows 95 RTM, Windows NT 4.0, Windows NT 3.51 and Windows 3.1 with Win32s, but it was not the case on NT 3.1 and 3.5, where the intersection was cross-shaped due to overshoot.
Here is a reference image generated on Windows NT 4.0, with the relevant pixels zoomed in:
And here is a screenshot of the overshoot on NT 3.5, again with the relevant pixels zoomed in:
The Polyline() function was also tested with the same results. Other GDI functions were not tested. As far as I can tell, this GDI quirk is, as-yet undocumented. I skimmed the BetaWiki page on GDI but could not find any mention of this.
According to the documentation of the Win32 LineTo() function, the line is drawn from the starting point inclusive to the ending point exclusive. This is true on Windows NT 3.51+, Windows 9x and even Win32s on Windows 3.1. On NT 3.1 and 3.5, however, the ending point is included. This can make GDI-drawn shapes look odd, as the lines may overshoot their intended connection point, especially for series of connected lines in different colours.
To test this, I wrote a Win32 program drawing a horizontal and vertical line. The horizontal line goes from point (11, 110) to (111 110) in the client area. The vertical line goes from (110, 11) to (110, 111) in the client area. Because GDI is not supposed to include the end point, and as only one coordinate changes per-pixel for each line, the last pixel of each line should be located at (110, 110), resulting in a perfect corner. Indeed, this was the case on Windows 95 RTM, Windows NT 4.0, Windows NT 3.51 and Windows 3.1 with Win32s, but it was not the case on NT 3.1 and 3.5, where the intersection was cross-shaped due to overshoot.
Here is a reference image generated on Windows NT 4.0, with the relevant pixels zoomed in:
And here is a screenshot of the overshoot on NT 3.5, again with the relevant pixels zoomed in:
The Polyline() function was also tested with the same results. Other GDI functions were not tested. As far as I can tell, this GDI quirk is, as-yet undocumented. I skimmed the BetaWiki page on GDI but could not find any mention of this.