Tcl Source Code

I found very weird discrepancy regarding Tcl_FSStat using 8.6 under Windows for file sub-commands (that use GetStatBuf), depending from used path by "built-in Windows names with dir path and extension".

The explanation, what happens, could be provided from this script:

proc test_fs_nef {{path {}} {fn this-file-does-not-exists.txt}} {
  puts "[set fn [file join $path $fn]] ..."
  puts -nonewline [format [string repeat "  %-8s: \[%d\] %s\n" 5] \
    atime [catch {file atime $fn} r] $r \
    mtime [catch {file mtime $fn} r] $r \
    size  [catch {file size $fn} r] $r \
    stat  [catch {file stat $fn stat} r] "$r [array get stat]" \
    type  [catch {file type $fn} r] $r \
  ]
}

puts [test_fs_nef [info library]]
puts [test_fs_nef [info library] CON.txt]

puts [test_fs_nef $::env(TEMP)]
puts [test_fs_nef $::env(TEMP) CON.txt]; ## here something goes broken!!!

D:/Projects/tcl8.6-upstream/library/this-file-does-not-exists.txt ...
  atime   : [1] could not read "D:/Projects/tcl8.6-upstream/library/this-file-does-not-exists.txt": no such file or directory
  mtime   : [1] could not read "D:/Projects/tcl8.6-upstream/library/this-file-does-not-exists.txt": no such file or directory
  size    : [1] could not read "D:/Projects/tcl8.6-upstream/library/this-file-does-not-exists.txt": no such file or directory
  stat    : [1] could not read "D:/Projects/tcl8.6-upstream/library/this-file-does-not-exists.txt": no such file or directory
  type    : [1] could not read "D:/Projects/tcl8.6-upstream/library/this-file-does-not-exists.txt": no such file or directory

D:/Projects/tcl8.6-upstream/library/CON.txt ...
  atime   : [1] could not read "D:/Projects/tcl8.6-upstream/library/CON.txt": no such file or directory
  mtime   : [1] could not read "D:/Projects/tcl8.6-upstream/library/CON.txt": no such file or directory
  size    : [1] could not read "D:/Projects/tcl8.6-upstream/library/CON.txt": no such file or directory
  stat    : [1] could not read "D:/Projects/tcl8.6-upstream/library/CON.txt": no such file or directory
  type    : [1] could not read "D:/Projects/tcl8.6-upstream/library/CON.txt": no such file or directory

C:/Temp/this-file-does-not-exists.txt ...
  atime   : [1] could not read "C:/Temp/this-file-does-not-exists.txt": no such file or directory
  mtime   : [1] could not read "C:/Temp/this-file-does-not-exists.txt": no such file or directory
  size    : [1] could not read "C:/Temp/this-file-does-not-exists.txt": no such file or directory
  stat    : [1] could not read "C:/Temp/this-file-does-not-exists.txt": no such file or directory
  type    : [1] could not read "C:/Temp/this-file-does-not-exists.txt": no such file or directory

C:/Temp/CON.txt ...
  atime   : [1] could not get access time for file "C:/Temp/CON.txt"
  mtime   : [1] could not get modification time for file "C:/Temp/CON.txt"
  size    : [0] 0
  stat    : [0]  mtime 0 atime 0 gid 0 nlink 0 mode 8630 type characterSpecial ctime 0 uid 0 ino 0 size 0 dev -1
  type    : [0] characterSpecial

The guess that something may be wrong on Windows self (e. g. other return code from WinAPI) has not been confirmed: the same script executed for 8.5th returns the expected results in the last case (exactly as for all 3 first cases). It's not the path and file-name alone, because this weird behavior happens only by the combination temp-path together with "CON.txt".

This reflects why the several cmdAH-test-cases fail if used without -tmpdir and successful with -tmpdir parameter.

.\tclsh86 ..\..\tests\all.tcl -file "cmdAH.*"
  cmdAH-20.7.1 FAILED
  cmdAH-24.14.1 FAILED
  cmdAH-27.4.1 FAILED
  cmdAH-28.13.1 FAILED
  cmdAH-29.6.1 FAILED
Total   320     Passed  140     Skipped 175     Failed  5

.\tclsh86 ..\..\tests\all.tcl -file "cmdAH.*" -tmpdir %TEMP%
Total   320     Passed  145     Skipped 175     Failed  0

fixed in [d61a95dfb6], this close.

Yep that was the reason.
But I found the difference - in case of con.txt in temp-folder the NativeStat does open it with CreateFile (but GetFileInformationByHandle fails) and GetFileType returns 2 (FILE_TYPE_CHAR), so ignored in 8.6 (but returns ENOENT in 8.5).

So we can ignore it (as weird Windows hack) and I'll repair the test-cases now.

It was two years ago so most likely it would have been either XP or Windows 8 I suspect. I don't think I'd moved to Windows 10 by that time.

Also, rethinking the issue I'm wondering whether rather than being pedantic, from a practical point of view it would have been better to have returned 0 as the time instead of raising an error. I still don't feel good about "lying" about the timestamp like that though. Also, I don't think that it is relevant to this specific ticket since you are addressing why the behaviour is *different* in the two cases as opposed to what the specific behaviour is.

/Ashok

@Ashok, you've certainly overseen, that all this the test-cases you've made, fail only in if no tmpdir specified (see again the ticket description about the combination where it has different behavior). Neither the test-cases are changed, nor something else (I tested your commit [5e0bded748], it fails likewise if no temp-dir specified.)

Which windows-version do you used?

https://core.tcl.tk/tcl/tktview?name=ae61a67192 is the specific ticket. Change was made in July 2016.

From the documentation - " If the file does not exist or its access time cannot be queried or set then an error is generated."

For the built-in devices like CON on Windows, access and modification times are not maintained. I had changed the code to raise an error accordingly as logged in some old ticket.

However, I am fairly certain all tests passed at the time I made the change (it was a while ago) so either these are new tests or something else has changed.

/Ashok

Maybe some special case processing for the CON: device
that has gone wrong???