<div dir="ltr"><div class="gmail_extra"><div class="gmail_quote">On Tue, Aug 4, 2015 at 2:39 AM, Daniel Vetter <span dir="ltr"><<a href="mailto:daniel@ffwll.ch" target="_blank">daniel@ffwll.ch</a>></span> wrote:<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Doesn't this have the effect of accidentally removing old results you<br>
might want to keep if you do a type (with e.g. shell completion)? Old<br>
piglit refused to run if the results dir was there iirc, so maybe we need<br>
a --force-overwrite option or similar?<br>
-Daniel<br></blockquote><div><br></div><div>For summary it does fail if the results already exist, but for run I don't think it ever has required --overwrite option. This is a bug of the transition from having a single open file to using one file per test to atomicity in piglit.</div><div><br></div><div>Since we moved to JSON, the results file would be opened, each test would be written into it serially, and at the end it would be closed. This leads to some odd bugs if you do something like this:</div><div>`piglit run quick deqp-gles3 foo -c`</div><div><br></div><div>let it run, so 35,000 tests, and then stop it with ctrl+\, and then do:</div><div>`piglit run quick foo -c`</div><div><br></div><div>Since the quick profile only as ~25000 tests IIRC, you'll end up with up to 10,000 tests from the original run scooped up into the new run, leading to very odd bugs. So we remove then at the start. In fact, the only way these files should be used is during a `piglit resume`</div><div><br></div><div>This doesn't change the fact that results.json will be overwritten if you do this. If we want to have a --overwrite switch we can do that, but that's a separate change.</div><div><br></div><div>Does that make sense? Do I just need to change the commit message to be clearer?</div><div><br></div><div>[snip]</div></div></div></div>