There is a (Dutch parliament) document linked on this page https://www.tweedekamer.nl/kamerstukken/amendementen/detail?id=2020Z01578&did=2020D03316 Downloads as Amendement_van_het_lid_Verhoeven_34972-23_over_de_erkenning_langs_de_kamers_voor_verlening.dot Try to open in ~master > error Try to open in 6.0.7.2 > Opens fine Rename to .doc > opens in master.
On Win10 with master sources updated today, I could reproduce this.
Regression from: https://cgit.freedesktop.org/libreoffice/core/commit/?id=65559252f138aada7a55d3c5fe0a932a222d13e0 tdf#110987: Don't mis-detect .doc files as .dot Also add a unit test for that. Change-Id: I86c195cebbe12b2bdf498954956db882f6f0d12b Reviewed-on: https://gerrit.libreoffice.org/68005 Tested-by: Jenkins Reviewed-by: Tor Lillqvist <tml@collabora.com> Indeed, reverting this patch makes LO able to open dot file. Tor: one for you?
Now we just have to chose which misbehaviour is the best one. I know, let's switch back and forth in each major version! (Joke.) Or let's pop up a dialog asking the user! (Also a joke.) Seriously: No, I don't want to touch this any more. Anybody is free to revert my change, of course. Why would I care.
s/chose/choose
Tor: just to be sure my previous comment was misinterpreted, when I told regression from ..., I didn't mean : "you made the commit, you're the culprit, you must fix this". I just meant this patch triggered this regression but perhaps the patch revealed another cornercase bug in addition with fixing tdf#110987 and maybe you may have some idea since you had worked on it. Of course, don't hesitate to uncc yourself. About the bug itself, I'm giving a look at https://interoperability.blob.core.windows.net/files/MS-DOC/%5bMS-DOC%5d-190319.pdf, trying to understand how FibBase is built.
Cor: just for the test, I opened the file on Word 365 and saved it (without any changed on a new file). Here's hexdump of extract FibBase struct(see my previous comment) before: EC A5 C1 00 63 00 13 04 00 00 F0 Here's hexdump of extract FibBase struct after: EC A5 C1 00 6F 00 09 04 00 00 F1 "F1" = 11110001 So "fDot" (the bit tested in Tor's patch) = 1, here (contrary to the initial doc) and the new dot file can be opened on LO. I think I already read something about MsOffice file validation but not sure. The idea would be to validate dot file provided by the website.
Created attachment 157564 [details] dot file resaved Here's the dot file resaved which is correctly opened on LO.
This is the same as bug 125216 if I'm not mistaken. *** This bug has been marked as a duplicate of bug 125216 ***