Allow NVDA to still function if a component changes the current directory #11650

michaelDCurran · 2020-09-21T02:37:09Z

Link to issue number:

Summary of the issue:

When testing the Ivona SAPI5 demo voices for another issue from https://www.harposoftware.com/ I noticed that once loading one of these voices:

NVDA could no longer save its configuration
Could no longer switch back to eSpeak or OneCore
If restarted with this voice, it could no longer speak at all, due to NVDA being unable to load character descriptions / symbol data.

All of this is caused by the fact that this SAPI5 engine very inappropriately changes the current directory to its own directory when the voice is loaded, and never changes it back.
Many parts of NVDA assume that the current directory is set to NVDA's application directory, and therefore all relative paths (and there are many of them) depend on the current directory.

As much as this engine is behaving very badly, we should at very least make sure that NVDA does not become unusable. I.e. config should still be able to be saved, symbols (and therefore speech) should still work, and switching back to another synthesizer should also work.

Description of how this pull request fixes the issue:

When NVDA first executes, as well as changing the current directory to the application dir, also save this application dir off to globalVars.appDir so that other parts of NVDA can use it later on.
ensure that all paths given as commandline arguments are made absolute as soon as possible, before the current directory has a chance to change to something we did not expect.
Replace the majority of relative paths with absolute paths made from globalVars.appDir, rather than just calling os.path.abspath. This includes the config dir, eSpeak paths, nvdaHelper dll paths, gui doc file paths, and more.

It is possible that some paths have been missed, but with these changes and running SAPI5 Ivona it is certainly now possible to:

save configuration
Switch to eSpeak or Onecore
speak with correct symbols after restarting.

Testing performed:

See the three points above.

Known issues with pull request:

None

Change log entry:

When using the SAPI5 Ivona voices from harposoftware.com, NvDA is now able to save configuration, switch synthesizers, and no longer will stay silent after restarting.

… made absolute as soon as possible to protect against the current directory changing later on. Also store NVDA's app dir in globalVars.

…aths.

source/logHandler.py

lukaszgo1 · 2020-09-21T08:49:13Z

There are indeed some places where NVDA expects current directory to be its install dir which you've missed:

in logHandler _getDefaultLogFilePath writes log to the current dir if running from sources.
TABLES_DIR in brailletables
Path to the html template in ui browseableMessage
Most places where we call nvwave.playWaveFile

Please also grep for places in which we call os.getcwd

…ars.appDir calls.

…ttext.translate

…ory.

… file.

AppVeyorBot · 2020-09-22T21:13:05Z

Build (for testing PR)

See test results for failed build of commit 42151e4275

lukaszgo1 · 2020-09-22T23:12:44Z

Two more cases of relative path's are SLAVE_FILENAME in config and path to the lilli.dll in brailleDisplayDrivers. Also this PR fixes issue #6491

lukaszgo1 · 2020-09-23T08:53:13Z

Sorry one more - path to builtin.dic in speechDictHandler is also relative.

…n places where we can't be completely sure the path is absolute.

source/addonHandler/__init__.py

lukaszgo1 · 2020-09-24T19:59:55Z

source/brailleDisplayDrivers/lilli.py

 from logHandler import log
+from ctypes import windll
 from ctypes import *


windll is the only part of ctypes used in this file - this line can be removed.

lukaszgo1 · 2020-09-24T20:07:32Z

source/nvda_slave.pyw

+			h = ctypes.windll.kernel32.LoadLibraryExW(
+				os.path.join(globalVars.appDir, "lib", versionInfo.version, "nvdaHelperRemote.dll"),
+				0,
+				0x8


Please make this a named constant in winKernel and use it both here and in NVDAHelper.

…in NVDAHelper and nvda_slave when loading NvDAHelperRemote.

codeofdusk · 2020-09-26T23:37:21Z

Do you plan to merge this fairly quickly? (i.e. does it make sense to change #11639 so it uses globalVars.appDir, therefore blocking that PR on this one)?

michaelDCurran · 2020-09-29T00:55:21Z

@codeofdusk This pr may take a little time to be carefully reviewed I think.

codeofdusk · 2020-09-29T00:57:41Z

OK, I'll leave #11639 as-is (to avoid unnecessary delay) but we should make a note to adjust the nvda_dmp invocation to use globalVars.appDir once this goes in.

feerrenrut

looks good to me @michaelDCurran

josephsl · 2020-09-30T00:53:23Z

Hi,

I'm afraid this PR introduced startup regression - the resulting alpha build isn't even starting at all, or if it did, instlal screen or UAC (in case of an update) isn't showing.

Thanks.

michaelDCurran · 2020-09-30T00:56:56Z

Most likely nvdaRemote. I am investigating why this is the case right now. @josephsl can you confirm that if you do have NVDARemote installed, disabling / removing allows NVDA to start again? or is it other add-ons as well.

michaelDCurran · 2020-09-30T01:04:10Z

I have temporarily reset master to the commit before this merge while we work out why this causes an incompatibility with NVDARemote and thus stops NVDA from starting.

michaelDCurran · 2020-09-30T01:19:44Z

See NVDARemote/NVDARemote#225

michaelDCurran · 2020-09-30T01:39:45Z

NVDA is stuck in a while loop.
The stack for the main thread is:

Python stack for thread 13648 (MainThread):
  File "source/nvda.pyw", line 236, in <module>
  File "C:\Users\mick\programming\git\nvda\source\core.py", line 483, in main
    globalPluginHandler.initialize()
  File "C:\Users\mick\programming\git\nvda\source\globalPluginHandler.py", line 32, in initialize
    for plugin in listPlugins():
  File "C:\Users\mick\programming\git\nvda\source\globalPluginHandler.py", line 23, in listPlugins
    plugin = importlib.import_module("globalPlugins.%s" % name, package="globalPlugins").GlobalPlugin
  File "C:\Program Files (x86)\Python37-32\lib\importlib\__init__.py", line 127, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 1006, in _gcd_import
  File "<frozen importlib._bootstrap>", line 983, in _find_and_load
  File "<frozen importlib._bootstrap>", line 967, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 677, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 728, in exec_module
  File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
  File "C:\Users\mick\programming\git\nvda\source\userConfig\addons\nvdaRemote\globalPlugins\remoteClient\__init__.py", line 25, in <module>
    addonHandler.initTranslation()
  File "C:\Users\mick\programming\git\nvda\source\addonHandler\__init__.py", line 533, in initTranslation
    addon = getCodeAddon(frameDist=2)
  File "C:\Users\mick\programming\git\nvda\source\addonHandler\__init__.py", line 525, in getCodeAddon
    while curdir not in _getDefaultAddonPaths():
  File "C:\Users\mick\programming\git\nvda\source\addonHandler\__init__.py", line 176, in _getDefaultAddonPaths
    return addon_paths

…g them to stop an infinite while loop introduced in #11650

michaelDCurran · 2020-09-30T03:04:45Z

in addonHandler.getCodeAdon: we walk back up to the root-most directory of an addon, but we need to normalize the paths with os.path.normpath before comparing them. Previously os.path.abspath did this for us.

michaelDCurran · 2020-09-30T03:06:26Z

Leaving master reset back to the commit before this (I.e. this pr has been reverted). I shall open a new pr containing all the current changes plus the fix.

…tory (#11707) * nvda.pyw: ensure that all paths coming from commandline arguments are made absolute as soon as possible to protect against the current directory changing later on. Also store NVDA's app dir in globalVars. * Use the NVDA app dir rather than the current directory for relative paths. * Fix unit tests. * Remove all usage of os.getcwd and replace it with globalVars.appDir * Replace all remaining os.path.join("* calls with os.path.join(globalVars.appDir calls. * nvda.pyw: provide an absolute path to gettext.translate * nvda_slave: set globalVars.appDir, and provide an absolute path to gettext.translate * getDefaultLogFilePath no longer uses the current directory. * brailleTables: TABLES_DIR is no longer relative to the current directory. * ui.browsableMessage no longer uses a relative path to get to the html file. * Change all playWavefile calls to be non-relative * Fix linting issues * another relative wave file path * Fix linting issues * speechDictHandler: the path to builtin.dic is no longer relative. * config: slave_fileName is no longer relative * Lilli braille driver: path to dll is no longer relative. * Fix linting issues * nvda_slave: don't load nvdaRemote with a relative path. * Remove all usage of os.path.abspath, but add a couple of assertions in places where we can't be completely sure the path is absolute. * Fix translation comments * Add the ALTERED_LIBRARY_SEARCH_PATH constant to winKernel and use it in NVDAHelper and nvda_slave when loading NvDAHelperRemote. * Lili braille dirver: remove unneeded import. * Update what's new * addonHandler.getCodeAddon: make sure to normalize paths when comparing them to stop an infinite while loop introduced in #11650

michaelDCurran added 2 commits September 21, 2020 11:24

nvda.pyw: ensure that all paths coming from commandline arguments are…

817c4ca

… made absolute as soon as possible to protect against the current directory changing later on. Also store NVDA's app dir in globalVars.

Use the NVDA app dir rather than the current directory for relative p…

cd8e8d6

…aths.

This comment has been minimized.

Sign in to view

Fix unit tests.

ffcffc9

This comment has been minimized.

Sign in to view

lukaszgo1 reviewed Sep 21, 2020

View reviewed changes

source/logHandler.py Show resolved Hide resolved

michaelDCurran added 2 commits September 22, 2020 14:24

Remove all usage of os.getcwd and replace it with globalVars.appDir

347f5c4

Replace all remaining os.path.join("* calls with os.path.join(globalV…

6026250

…ars.appDir calls.

This comment has been minimized.

Sign in to view

michaelDCurran added 2 commits September 22, 2020 21:31

nvda.pyw: provide an absolute path to gettext.translate

879960c

nvda_slave: set globalVars.appDir, and provide an absolute path to ge…

8bcb481

…ttext.translate

This comment has been minimized.

Sign in to view

michaelDCurran added 5 commits September 22, 2020 23:14

getDefaultLogFilePath no longer uses the current directory.

e8f840d

brailleTables: TABLES_DIR is no longer relative to the current direct…

fb4affc

…ory.

ui.browsableMessage no longer uses a relative path to get to the html…

dcecdf2

… file.

Change all playWavefile calls to be non-relative

102c68e

Fix linting issues

1da2d23

michaelDCurran added 2 commits September 23, 2020 08:03

another relative wave file path

b1850be

Fix linting issues

d5501bf

This comment has been minimized.

Sign in to view

michaelDCurran added 6 commits September 23, 2020 21:18

speechDictHandler: the path to builtin.dic is no longer relative.

631782d

config: slave_fileName is no longer relative

8ad7bfc

Lilli braille driver: path to dll is no longer relative.

bc49fe7

Fix linting issues

0413ad0

nvda_slave: don't load nvdaRemote with a relative path.

bf4b1b8

Remove all usage of os.path.abspath, but add a couple of assertions i…

3be773c

…n places where we can't be completely sure the path is absolute.

lukaszgo1 reviewed Sep 24, 2020

View reviewed changes

Merge branch 'master' into fixAbspathCalls

495d164

This comment has been minimized.

Sign in to view

Add the ALTERED_LIBRARY_SEARCH_PATH constant to winKernel and use it …

3c0f741

…in NVDAHelper and nvda_slave when loading NvDAHelperRemote.

michaelDCurran force-pushed the fixAbspathCalls branch from 7e76868 to 3c0f741 Compare September 25, 2020 03:23

Lili braille dirver: remove unneeded import.

6000aec

This comment has been minimized.

Sign in to view

feerrenrut previously approved these changes Sep 29, 2020

View reviewed changes

michaelDCurran added 2 commits September 30, 2020 08:30

Merge branch 'master' into fixAbspathCalls

4dc3192

Update what's new

a9baa8c

michaelDCurran dismissed feerrenrut’s stale review via a9baa8c September 29, 2020 22:36

michaelDCurran merged commit ba42885 into master Sep 29, 2020

michaelDCurran deleted the fixAbspathCalls branch September 29, 2020 23:13

nvaccessAuto added this to the 2020.4 milestone Sep 29, 2020

codeofdusk mentioned this pull request Sep 30, 2020

add-on breaks NVDA initialization NVDARemote/NVDARemote#225

Closed

michaelDCurran added a commit that referenced this pull request Sep 30, 2020

addonHandler.getCodeAddon: make sure to normalize paths when comparin…

2ae3811

…g them to stop an infinite while loop introduced in #11650

michaelDCurran mentioned this pull request Sep 30, 2020

pr11650 with fix for addon paths #11707

Merged

michaelDCurran mentioned this pull request Oct 21, 2020

languageHandler.getAvailableLanguages: use globalVars.appDir rather than depending on the current directory #11778

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Allow NVDA to still function if a component changes the current directory #11650

Allow NVDA to still function if a component changes the current directory #11650

michaelDCurran commented Sep 21, 2020 •

edited

Loading

This comment has been minimized.

This comment has been minimized.

lukaszgo1 commented Sep 21, 2020

This comment has been minimized.

This comment has been minimized.

AppVeyorBot commented Sep 22, 2020

This comment has been minimized.

lukaszgo1 commented Sep 22, 2020

lukaszgo1 commented Sep 23, 2020

lukaszgo1 Sep 24, 2020

lukaszgo1 Sep 24, 2020

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

codeofdusk commented Sep 26, 2020

michaelDCurran commented Sep 29, 2020

codeofdusk commented Sep 29, 2020

feerrenrut left a comment

josephsl commented Sep 30, 2020

michaelDCurran commented Sep 30, 2020 via email

michaelDCurran commented Sep 30, 2020

michaelDCurran commented Sep 30, 2020

michaelDCurran commented Sep 30, 2020

michaelDCurran commented Sep 30, 2020

michaelDCurran commented Sep 30, 2020

Allow NVDA to still function if a component changes the current directory #11650

Allow NVDA to still function if a component changes the current directory #11650

Conversation

michaelDCurran commented Sep 21, 2020 • edited Loading

Link to issue number:

Summary of the issue:

Description of how this pull request fixes the issue:

Testing performed:

Known issues with pull request:

Change log entry:

This comment has been minimized.

This comment has been minimized.

lukaszgo1 commented Sep 21, 2020

This comment has been minimized.

This comment has been minimized.

AppVeyorBot commented Sep 22, 2020

This comment has been minimized.

lukaszgo1 commented Sep 22, 2020

lukaszgo1 commented Sep 23, 2020

lukaszgo1 Sep 24, 2020

Choose a reason for hiding this comment

lukaszgo1 Sep 24, 2020

Choose a reason for hiding this comment

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

codeofdusk commented Sep 26, 2020

michaelDCurran commented Sep 29, 2020

codeofdusk commented Sep 29, 2020

feerrenrut left a comment

Choose a reason for hiding this comment

josephsl commented Sep 30, 2020

michaelDCurran commented Sep 30, 2020 via email

michaelDCurran commented Sep 30, 2020

michaelDCurran commented Sep 30, 2020

michaelDCurran commented Sep 30, 2020

michaelDCurran commented Sep 30, 2020

michaelDCurran commented Sep 30, 2020

michaelDCurran commented Sep 21, 2020 •

edited

Loading