@@ -441,6 +441,10 @@ preference, or `~/.ECCE/RemoteServer/`, which `ecce-remote-setup <host>
 --user --auto` writes without root) and changes from **Edit > Change
 Server...** in the Organizer, which shows the current mode and data folder.
 The change applies at the next start; nothing is copied between modes.
+The data server's login window has a button **Use this computer
+instead...** that does the same as choosing this computer there
+(`ecce-first-start --apply local`) and ends that start with a message to
+start again.
 On Windows `ecce.cmd` runs the window with the package's own Python and, for
 a server, starts the session the same way `ecce -remote` does (the broker
 file, then the Organizer). On macOS the window needs wxPython, which
@@ -463,10 +467,13 @@ asked** where an administrator or the command line decides:
 `siteconfig/RemoteServer` present (an administrator or `ecce-remote-setup` as
 root), `ecce -remote` or `ECCE_REMOTE_SERVER`, `ECCE_LOCAL_DATA` or `ecce
 --local`, a shared broker declared (`siteconfig/SharedBroker`), an account
-marked as a central server (`~/.ECCE/mosquitto.server`), and on Linux **any
-installation that has the `ecce-server` package** (the server itself, and
-remote desktop sessions such as FastX on it). So modes 2 and 3, the server
-machine and the all-in-one install behave exactly as before.
+marked as a central server (`~/.ECCE/mosquitto.server`). An installed
+`ecce-server` package is not configuration: its users are asked too, and a
+user who already works on the per-user data server finds **Store data on
+this computer** preselected, where Continue keeps that data server. On a
+server whose users log in to it (FastX), `ecce-remote-setup` run as root
+(`siteconfig/RemoteServer`) points everyone at the server and nothing is
+asked.
 `ECCE_NO_FIRST_START=1` never asks (tests and scripts set it), and without a
 display or wxPython nothing is asked.
 
 
@@ -0,0 +1,26 @@
+---
+type: pitfall
+title: "genmol is handed the whole molecule after the Builder's Find"
+area: codereg
+paths: ["src/apps/symmetry/dosymops.f", "src/apps/symmetry/genmol.f", "src/viz/sgcommands/FindSymmetryCmd.C", "src/tdat/chemistry/Fragment.C", "src/apps/calced/CalcEd.C", "src/apps/calced/ESInputController.C"]
+issues: []
+---
+**Symmetry > Find sets `Fragment::useSymmetry(true)` on the whole molecule,
+not on its symmetry-unique atoms.** The Calculation Editor (and the input
+writer's `NumElectrons`) then run `Fragment::generateFullMolecule()`, i.e.
+genmol, on every atom. genmol must therefore return a molecule that is
+already complete unchanged.
+
+It did not: in `dosymops.f` the "is this atom already present" test had an
+empty body, so each atom's images were added again. Water came back as
+H4O (5 atoms, 12 electrons), methane as 17 atoms, benzene as 72, in every
+release from the PNNL base to 9.0.0-alpha.8. The test now skips an atom
+already generated from an earlier one; from an irreducible set the result
+is unchanged. Only for molecules (`itype` 0): genmollat shares
+`dosymops.f`, works in fractional coordinates, and the Builder's lattice
+tools (Generate, supercells, folding; `tests/apps/pbc_edit_test.py`,
+`nmode_edit_test.py`) lost atoms when the skip applied there too.
+
+Tests: `tests/symmetry/run_tests.py` (`checkGenmol`: full and irreducible
+input for water, methane, benzene) and `tests/apps/symmetry_find_test.py`
+(the real Builder Find and Calculation Editor).
@@ -120,6 +120,7 @@ unit is produced in `ai.<code>`/the `.tpl`, closest to the input deck.
 - [Never point an `ai.<code>` script's `-t` at the repo's own template](never-point-an-ai-code-scripts-t.md)
 - [Codereg dialogs are never told which elements the structure contains](codereg-dialogs-are-never-told-which-elements.md)
 - [CalcEd's own ES.Theory.UseSymmetry key is lost whenever p_GUIValues is replaced](calced-owns-use-symmetry-key.md)
+- [genmol is handed the whole molecule after the Builder's Find, and must return it unchanged](genmol-is-handed-the-whole-molecule.md)
 - [A dialog's choice string must match the generator's expected string EXACTLY, or selecting it silently emits nothing](a-dialogs-choice-string-must-match-the.md)
 - [The basis-set writers decide *how* a basis reaches the deck, and every bug in them is silent](the-basis-set-writers-decide-how-a.md)
 - [A Gaussian basis keyword is selectable only if the library has an entry; tools/basissets/g16_basis_dump.py makes one from the keyword](a-gaussian-basis-keyword-needs-a-library-entry.md)
 
@@ -11,7 +11,8 @@ asks once per user ("Store data on this computer" / "Connect to a server"); the
 answer goes to `~/.ECCE/first-start-answer`. Only that answer and explicit
 configuration skip it (`skip_reason()`: siteconfig/RemoteServer,
 SharedBroker, `-remote`, `--local`/ECCE_LOCAL_DATA, a central-server account,
-the server package on Linux, no display, ECCE_NO_FIRST_START). Existing data
+no display, ECCE_NO_FIRST_START). An installed server package is not
+configuration and does not skip it (Andy, 2026-10-09). Existing data
 (a local folder, ~/.ECCE/dataserver, a server chosen, the preference) only
 preselects the window, which then shows the current setup; keeping a
 per-user data server records the answer without switching the preference.
@@ -52,6 +53,17 @@ to `tests/apps/first_start_test.py`.
   Edit > Change Server... starts the script with `python3w.exe` on Windows.
 - macOS: ECCE.app bundles no Python or wxPython, so the window is skipped
   silently (ImportError) until it does; local data stays the default there.
-- Tests set `ECCE_FIRST_START_ANSWER` (`local`, `server:host[:port]`) and
-  `ECCE_FIRST_START_SHOT`; never synthetic input. `tests/apps/session_end.py
+- The data server login window (WxAuth, shown by `WxDavAuth::prompt` in
+  a data-server session) has "Use this computer instead...": it runs
+  `ecce-first-start --apply local`, says the change applies at the next
+  start, and ends this one (the gateway skips its "Authentication
+  Failure" message, other apps leave their main loop). Restart rather
+  than switch in place: the session's broker and -remote state are
+  already set up. Test hook `ECCE_TEST_AUTH_ANSWER=uselocal[:s]`;
+  `tests/apps/session_end.py auth-uselocal`.
+- Tests set `ECCE_FIRST_START_ANSWER` (`local`, `server:host[:port]`,
+  `continue` = Continue on what is preselected; the window logs
+  "preselected local|server") and `ECCE_FIRST_START_SHOT`; never synthetic
+  input. The suite's isolated user (`tests/apps/isolate.py`) has a recorded
+  answer, so the other cases are not asked. `tests/apps/session_end.py
   first-local first-server` drive it through the real `ecce`.
@@ -58,3 +58,5 @@ what the deployment mode needs (GETTING_STARTED, "Deployment modes"):
 - [macOS: ECCE.app carries its own Python with wxPython; every python3 ECCE runs must find it first (#133)](macos-theory-dialogs-need-python3-with-wxpython.md)
 - [macOS: BSD env has no -0, so a machine's login file was silently ignored (#133)](macos-env-has-no-dash-zero.md)
 - [tests/macos/e2e.sh: water single points on a Mac with real codes](macos-end-to-end-run.md)
+- [Windows: the user's folder can hold a space, and every path ECCE builds sits below it (#247)](windows-paths-hold-spaces.md)
+- [Windows: a text-mode stream's tellg() is not the length of what it reads; DAV PUT 400 (#247)](windows-dav-put-text-mode.md)
@@ -0,0 +1,24 @@
+---
+type: pitfall
+title: "Windows: a text-mode stream's tellg() is not the length of what it reads (DAV PUT 400)"
+area: services
+paths: ["src/dsm/dav/EcceDAVClient.C", "src/apps/calced/ESInputController.C", "src/comm/commxt/JobStore.C", "src/comm/commxt/Launch.C"]
+issues: [247]
+---
+**A Windows client saving to a data server got "input file copy to DAV
+failed" for every input deck** (Apache: 400, "An error occurred while
+reading the request body"). The deck is written by Strawberry Perl with
+CRLF line ends and handed to `putInputFile()` as a text-mode `ifstream`.
+`EcceDAVClient::getStreamSize()` took the Content-Length from
+`seekg(end)/tellg()`, which counts bytes on disk, while reading the
+stream drops each CR: the body came up short and the server gave up.
+Local mode never noticed, because it copies the file.
+
+On Windows the size is now counted by reading the stream, so what is sent
+is LF-only text and the length matches. Binary files must still be opened
+with `ios::binary` by the caller: a text-mode read also stops at a
+Ctrl-Z byte.
+
+Test: `tests/windows/central_win.py` (a Windows client against a central
+server: first-start answer, login through `-pipe`, a calculation made and
+its input saved and read back over WebDAV).
@@ -0,0 +1,34 @@
+---
+type: pitfall
+title: "Windows: the user's folder can hold a space, and every path ECCE builds sits below it"
+area: services
+paths: ["src/comm/rcommand/RCommand.C", "include/comm/RCommand.H", "src/comm/commxt/Launch.C", "src/comm/commxt/JobStore.C", "src/apps/calced/ESInputController.C", "src/apps/calced/CalcEd.C", "src/apps/calced/InputVerifier.C", "src/dsm/xml/JCode.C", "src/tdat/chemistry/SymmetryOps.C", "src/tdat/chemistry/Fragment.C", "scripts/gensub", "scripts/parsers/*.launchpp", "src/comm/rcommand/DirectTransportWin.C"]
+issues: [247]
+---
+**ECCE_HOME (`%LOCALAPPDATA%\Programs\ECCE`), the temp folder, the
+local data folder and the default run directory (`~/ecce-runs`) are all
+below `C:\Users\<name>`, and a Windows user name can contain a space.**
+Every command line that pastes one of these in unquoted breaks there,
+usually silently: Save said it could not create the temporary
+calculation directory (`test -d` on two words), the Theory/Runtype
+dialogs got their temp file as two arguments, gensub read the run
+directory up to the space from its param file, and a launchpp split
+`runDir: C:/...` at the drive colon.
+
+Rules: RCommand's file operations (`exists`, `directory`, `writable`,
+`cd`, `shellget`) take a *path*, not shell syntax; they quote it
+themselves. Command lines built elsewhere quote with
+`RCommand::quotePath()` (single quotes, a leading `~`/`~/` left bare so a
+remote shell still expands it) or `quoteGlob()` for a pattern; a line
+that may reach cmd.exe (`Ecce::runCommand`, `CreateProcess`) uses double
+quotes. gensub writes the run directory through `shWord()`, which leaves a
+plain path as it is, so the golden scripts did not change. A `key: value`
+reader splits at the first colon or space only.
+
+Test it with a tree *and* a state folder under a path with a space:
+`tests/windows/e2e_win.py "<tree with space>" "<state with space>"` (its
+temp folder is below the state folder); `tests/transport/testRCommandDirect.C`
+(`spaceChecks`) and `tests/queues/run_tests.py` (`spacedRunDir`) cover the
+Linux side. An `msiexec /a` tree has no `tmp` folder (the installer creates
+it), and MSYS then prints "could not find /tmp" on every shell start; make
+one when testing from such a tree.
@@ -119,9 +119,10 @@ Click **Continue** to confirm. ECCE remembers your choice and does not ask
 again. **Quit** closes ECCE without choosing anything.
 
 ECCE asks on Linux, macOS and Windows alike: on your own laptop you can
-connect to your teacher's server in the same way. It does not ask if it is
-already set up: on a computer that has a server configured by its
-administrator, on a server itself, or when you have used ECCE here before.
+connect to your teacher's server in the same way. Once you have answered,
+it does not ask again. It does not ask at all on a computer whose
+administrator has already set up a server, or in the account that runs the
+server.
 
 If ECCE cannot connect, it tells you in the same window. Check the address
 and your network connection, then click **Continue** again.
@@ -133,6 +134,10 @@ certificate ever changes, ECCE does not log you in and says so; ask the
 server's administrator, then choose **Edit > Change Server...** and connect
 again.
 
+The login window for a server also has a button **Use this computer
+instead...**. It makes the same choice as **Store data on this computer**
+and ends this start; start ECCE again to work on this computer.
+
 To change your mind later, choose **Edit > Change Server...** in the
 Organizer. The window says where ECCE works now, and which folder holds your
 calculations if it is this computer. The change applies the next time you
 
@@ -193,6 +193,14 @@ public:
     static string localityNote(const string& machine, const string& remShell,
                                const string& userName="");
     static bool isSameDomain(const string& machine);
+
+    // A path as one shell word, for command lines built here and by callers:
+    // single-quoted, except that a leading ~ or ~/ stays bare so the remote
+    // shell still expands it.  Windows user folders contain spaces.
+    static string quotePath(const string& path);
+    // A pattern whose last component is a glob: the directory part quoted,
+    // the last component left to the shell.
+    static string quoteGlob(const string& pattern);
     static string whereami(void);
 
 private:
 
@@ -49,6 +49,9 @@ public:
   void setPasswordLabel(const string& passwordLabel);
   void setProtocol(const string& protocol);
   void showChangeBtn(const bool& show);
+  /** The "Use this computer instead" button; it ends the dialog with
+      ID_BUTTON_AUTH_USE_LOCAL. */
+  void showUseLocal(const bool& show);
 
   /** A line under the fields saying the connection is encrypted; "" hides it. */
   void setEncryption(const string& note);