On Mon, 13 Sept 2021 at 11:57, Johan Corveleyn <jcor...@gmail.com> wrote: > > On Tue, Sep 7, 2021 at 3:58 PM Johan Corveleyn <jcor...@gmail.com> wrote: > > > > On Tue, Feb 23, 2021 at 9:59 AM Mariusz W <mawa...@gmail.com> wrote: > > > On 2021/01/26 14:54:04, Johan Corveleyn <jcor...@gmail.com> wrote: > > > > On Thu, Dec 24, 2020 at 10:25 AM Johan Corveleyn <jcor...@gmail.com> > > > > wrote: > > > > > On Wed, Dec 23, 2020 at 8:17 PM William A Rowe Jr > > > > > <wr...@rowe-clan.net> wrote: > > > > > > > > > > > > Sorry, I'd been pounding on a beta deliverable, it's my bad. This > > > > > > coming week I have cycles to give > > > > > > back to APR, CMake and other projects neglected during this crunch > > > > > > period. I hope an end-of-year > > > > > > (or very early January) release will meet your goals.. > > > > > > > > > > Yes, thanks, and no problem. I think that might be just in time, so > > > > > I'm crossing my fingers :-). > > > > > > > > > > (as you might know, Subversion development is going quite slowly these > > > > > days, so it's hard to say when the cycles of the different volunteers > > > > > will align -- it might be end of this year still, but I'm guessing > > > > > it's more likely to be early January too I think ... we're progressing > > > > > with small steps here and there) > > > Hi, > > > Maybe problem is in this line 627 in filestat.c [1] (changed in revision > > > [2]) mainly by adding APR_FINFO_LINK in line 627 > > > > > > Diff to prev: [3] > > > I think that this change is causing that "if" branch is not executed (svn > > > is using APR_FINFO_LINK in call) and in "else" branch we get error from > > > FindFirstFileW [6] because root (e.g. after subst) is ending with "/" or > > > "\" as in example from [4]) > > > The error code 720002 from apr_stat means ERROR_FILE_NOT_FOUND (code 2 > > > (0x2) [5]) from FindFirstFileW plus some adding and multiplication > > > operations in apr layer (last digit in 720002 is code 2). > > > > > > [1] - > > > https://svn.apache.org/viewvc/apr/apr/branches/1.7.x/file_io/win32/filestat.c?view=markup#l627 > > > [2] - https://svn.apache.org/r1855950 > > > [3] - > > > https://svn.apache.org/viewvc/apr/apr/branches/1.7.x/file_io/win32/filestat.c?r1=1795930&r2=1855950&pathrev=1855950&diff_format=h > > > [4] - > > > https://lists.apache.org/thread.html/rd890d13b826e8cd7acaa96769e10e7143b7e35e11c99bcaa1a75d481%40%3Cdev.apr.apache.org%3E > > > [5] - > > > https://docs.microsoft.com/en-us/windows/win32/debug/system-error-codes--0-499- > > > [6] - > > > https://svn.apache.org/viewvc/apr/apr/branches/1.7.x/file_io/win32/filestat.c?view=markup#l642 > > > > > > Regards, > > > Mariusz > > > > Thanks for looking into this, Mariusz. > > > > I think you might be onto something. I intended to look closer into > > your diagnosis, and perhaps try to test if changing those lines back > > to the original would fix it. But I didn't get around to it, sorry. > > > > Now that there was talk of an APR 1.7.1 release on this list, it > > popped back into my memory :-). > > > > @apr devs: what do you think about Mariusz pointer towards the > > possible cause of what we're seeing with SVN on Windows with apr > > 1.7.0? And is there an easy fix (for instance, just reverting that > > single hunk), which has a chance of being accepted into 1.7.1 (if so, > > I'll try to test it on my svn-dev machine, where I could reproduce the > > issue)? IIUC, right now apr_stat does seem to error out on paths > > ending with "/" or "\" on Windows, if APR_FINFO_LINK is given (while > > APR 1.6 did return a sensible result in that case). > > > > To be clear, IIUC Mariusz is referring to this hunk in filestat.c in > > r1855950: > > > > [[[ > > @@ -590,7 +664,7 @@ APR_DECLARE(apr_status_t) apr_stat(apr_f > > rv = apr_filepath_root(&root, &test, APR_FILEPATH_NATIVE, pool); > > isroot = (root && *root && !(*test)); > > > > - if ((apr_os_level >= APR_WIN_98) && (!(wanted & > > APR_FINFO_NAME) || isroot)) > > + if ((apr_os_level >= APR_WIN_98) && (!(wanted & > > (APR_FINFO_NAME | APR_FINFO_LINK)) || isroot)) > > { > > /* cannot use FindFile on a Win98 root, it returns \* > > * GetFileAttributesExA is not available on Win95 > > ]]] > > Oops, I meant (or Mariusz meant) this hunk in filestat.c r1855950: > > [[[ > > @@ -555,7 +624,7 @@ APR_DECLARE(apr_status_t) apr_stat(apr_f > if ((rv = utf8_to_unicode_path(wfname, sizeof(wfname) > / sizeof(apr_wchar_t), fname))) > return rv; > - if (!(wanted & APR_FINFO_NAME)) { > + if (!(wanted & (APR_FINFO_NAME | APR_FINFO_LINK))) { > if (!GetFileAttributesExW(wfname, GetFileExInfoStandard, > &FileInfo.i)) > return apr_get_os_error(); > > ]]] > > And yes, if I revert that single hunk, the Subversion problem with > subst'ed drives on Windows (or working copies on drive roots) is gone! > > Of course, I have no idea what other effects this has, but just > confirming that taking another turn in the above conditional (like it > was before) makes apr_stat return the same (AFAICS) as in 1.6.5, for > substed drives or drive roots on Windows. >
Hi, The problem with apr_stat(APR_FINFO_LINK | APR_FINFO_MIN) should be fixed by r1896717 [1] in trunk. This fix also should resolve performance regression in apr_stat() in most common cases. I plan to backport this fix to APR 1.7.x at some point later. Testing and review will be much appreciated! :) [1] https://svn.apache.org/r1896717 -- Ivan Zhakov
