I'm running the latest version of Selenium WebDriver with Geckodriver. I want to prevent Selenium from creating temporary Firefox Profiles in the temporary files directory when launching a new instance of WebDriver. Instead I want to use the original Firefox Profile directly. This has double benefit. First, it saves time (it takes significant amount of time for the profile to be copied to the temporary directory). Second, it ensures that cookies created during session are saved to the original profile. Before Selenium started relying on Geckodriver I was able to solve this problem by editing the class FirefoxProfile.class in SeleniumHQ as seen below:
public File layoutOnDisk() {
File profileDir;
if (this.disableTempProfileCreation) {
profileDir = this.model;
return profileDir;
} else {
try {
profileDir = TemporaryFilesystem.getDefaultTmpFS().createTempDir("ABC", "XYZ");
File userPrefs = new File(profileDir, "user.js");
this.copyModel(this.model, profileDir);
this.installExtensions(profileDir);
this.deleteLockFiles(profileDir);
this.deleteExtensionsCacheIfItExists(profileDir);
this.updateUserPrefs(userPrefs);
return profileDir;
} catch (IOException var3) {
throw new UnableToCreateProfileException(var3);
}
}
}
This would stop Selenium from creating a temporary Firefox Profile when the parameter disableTempProfileCreation was set to true.
However, now that Selenium is being controlled by Geckodriver this solution no longer works as the creation (and launch) of Firefox Profile is controlled by Geckodriver.exe (which is written in Rust language). How can I achieve the same objective with Geckodriver? I don't mind editing the source code. I'm using Java.
Thanks
Important Update:
I would like to thank everyone for taking the time to respond to this question. However, as stated in some of the comments, the first 3 answers do not address the question at all - for two reasons. First of all, using an existing Firefox Profile will not prevent Geckodriver from copying the original profile to a temporary directory (as indicated in the OP and clearly stated by one or more of the commentators below). Second, even if it did it is not compatible with Selenium 3.0.
I'm really not sure why 3 out of 4 answer repeat the exact same answer with the exact same mistake. Did they read the question? The only answer the even attempts to address the question at hand is the answer by #Life is complex however it is incomplete. Thanks.
UPDATED POST 05-30-2021
This is the hardest question that I have every tried to answer on Stack Overflow. Because it involved the interactions of several code bases written in multiple languages (Java, Rust and C++). This complexity made the question potentially unsolvable.
My last crack at this likely unsolvable question:
Within the code in your question you are modifying the file user.js This file is still used by Selenium.
public FirefoxProfile() {
this(null);
}
/**
* Constructs a firefox profile from an existing profile directory.
* <p>
* Users who need this functionality should consider using a named profile.
*
* #param profileDir The profile directory to use as a model.
*/
public FirefoxProfile(File profileDir) {
this(null, profileDir);
}
#Beta
protected FirefoxProfile(Reader defaultsReader, File profileDir) {
if (defaultsReader == null) {
defaultsReader = onlyOverrideThisIfYouKnowWhatYouAreDoing();
}
additionalPrefs = new Preferences(defaultsReader);
model = profileDir;
verifyModel(model);
File prefsInModel = new File(model, "user.js");
if (prefsInModel.exists()) {
StringReader reader = new StringReader("{\"frozen\": {}, \"mutable\": {}}");
Preferences existingPrefs = new Preferences(reader, prefsInModel);
acceptUntrustedCerts = getBooleanPreference(existingPrefs, ACCEPT_UNTRUSTED_CERTS_PREF, true);
untrustedCertIssuer = getBooleanPreference(existingPrefs, ASSUME_UNTRUSTED_ISSUER_PREF, true);
existingPrefs.addTo(additionalPrefs);
} else {
acceptUntrustedCerts = true;
untrustedCertIssuer = true;
}
// This is not entirely correct but this is not stored in the profile
// so for now will always be set to false.
loadNoFocusLib = false;
try {
defaultsReader.close();
} catch (IOException e) {
throw new WebDriverException(e);
}
}
So in theory you should be able to modify capabilities.rs in the geckodriver source code. That file contains the temp_dir.
As I stated this in only a theory, because when I looked at the Firefox source, which has temp_dir spread throughout the code base.
ORIGINAL POST 05-26-2021
I'm not sure that you can prevent Selenium from creating a temporary Firefox Profile.
From the gecko documents:
"Profiles are created in the systems temporary folder. This is also where the encoded profile is extracted when profile is provided. By default geckodriver will create a new profile in this location."
The only solution that I see at the moment would require you modify the Geckodriver source files to prevent the creation of temporary folders/profiles.
I'm currently looking at the source. These files might be the correct ones, but I need to look at the source more:
https://searchfox.org/mozilla-central/source/browser/app/profile/firefox.js
https://searchfox.org/mozilla-central/source/testing/mozbase/mozprofile/mozprofile/profile.py
Here are some other files that need to be combed through:
https://searchfox.org/mozilla-central/search?q=tempfile&path=
This looks promising:
https://searchfox.org/mozilla-central/source/testing/geckodriver/doc/Profiles.md
"geckodriver uses [profiles] to instrument Firefox’ behaviour. The
user will usually rely on geckodriver to generate a temporary,
throwaway profile. These profiles are deleted when the WebDriver
session expires.
In cases where the user needs to use custom, prepared profiles,
geckodriver will make modifications to the profile that ensures
correct behaviour. See [Automation preferences] below on the
precedence of user-defined preferences in this case.
Custom profiles can be provided two different ways:
1. by appending --profile /some/location to the [args capability],
which will instruct geckodriver to use the profile in-place;
I found this question on trying to do this: how do I use an existing profile in-place with Selenium Webdriver?
Also here is an issue that was raised in selenium on Github concerning the temp directory. https://github.com/SeleniumHQ/selenium/issues/8645
Looking through the source of geckodriver v0.29.1 I found a file where the profile is loaded.
source: capabilities.rs
fn load_profile(options: &Capabilities) -> WebDriverResult<Option<Profile>> {
if let Some(profile_json) = options.get("profile") {
let profile_base64 = profile_json.as_str().ok_or_else(|| {
WebDriverError::new(ErrorStatus::InvalidArgument, "Profile is not a string")
})?;
let profile_zip = &*base64::decode(profile_base64)?;
// Create an emtpy profile directory
let profile = Profile::new()?;
unzip_buffer(
profile_zip,
profile
.temp_dir
.as_ref()
.expect("Profile doesn't have a path")
.path(),
)?;
Ok(Some(profile))
} else {
Ok(None)
}
}
source: marionette.rs
fn start_browser(&mut self, port: u16, options: FirefoxOptions) -> WebDriverResult<()> {
let binary = options.binary.ok_or_else(|| {
WebDriverError::new(
ErrorStatus::SessionNotCreated,
"Expected browser binary location, but unable to find \
binary in default location, no \
'moz:firefoxOptions.binary' capability provided, and \
no binary flag set on the command line",
)
})?;
let is_custom_profile = options.profile.is_some();
let mut profile = match options.profile {
Some(x) => x,
None => Profile::new()?,
};
self.set_prefs(port, &mut profile, is_custom_profile, options.prefs)
.map_err(|e| {
WebDriverError::new(
ErrorStatus::SessionNotCreated,
format!("Failed to set preferences: {}", e),
)
})?;
let mut runner = FirefoxRunner::new(&binary, profile);
runner.arg("--marionette");
if self.settings.jsdebugger {
runner.arg("--jsdebugger");
}
if let Some(args) = options.args.as_ref() {
runner.args(args);
}
// https://developer.mozilla.org/docs/Environment_variables_affecting_crash_reporting
runner
.env("MOZ_CRASHREPORTER", "1")
.env("MOZ_CRASHREPORTER_NO_REPORT", "1")
.env("MOZ_CRASHREPORTER_SHUTDOWN", "1");
let browser_proc = runner.start().map_err(|e| {
WebDriverError::new(
ErrorStatus::SessionNotCreated,
format!("Failed to start browser {}: {}", binary.display(), e),
)
})?;
self.browser = Some(Browser::Host(browser_proc));
Ok(())
}
pub fn set_prefs(
&self,
port: u16,
profile: &mut Profile,
custom_profile: bool,
extra_prefs: Vec<(String, Pref)>,
) -> WebDriverResult<()> {
let prefs = profile.user_prefs().map_err(|_| {
WebDriverError::new(
ErrorStatus::UnknownError,
"Unable to read profile preferences file",
)
})?;
for &(ref name, ref value) in prefs::DEFAULT.iter() {
if !custom_profile || !prefs.contains_key(name) {
prefs.insert((*name).to_string(), (*value).clone());
}
}
prefs.insert_slice(&extra_prefs[..]);
if self.settings.jsdebugger {
prefs.insert("devtools.browsertoolbox.panel", Pref::new("jsdebugger"));
prefs.insert("devtools.debugger.remote-enabled", Pref::new(true));
prefs.insert("devtools.chrome.enabled", Pref::new(true));
prefs.insert("devtools.debugger.prompt-connection", Pref::new(false));
}
prefs.insert("marionette.log.level", logging::max_level().into());
prefs.insert("marionette.port", Pref::new(port));
prefs.write().map_err(|e| {
WebDriverError::new(
ErrorStatus::UnknownError,
format!("Unable to write Firefox profile: {}", e),
)
})
}
}
After looking through the gecko source it looks like mozprofile::profile::Profile is coming from FireFox and not geckodriver
It seems that you might have issues with profiles when you migrate to Selenium 4.
ref: https://github.com/SeleniumHQ/selenium/issues/9417
For Selenium 4 we have deprecated the use of profiles as there are other mechanisms that we can do to make the start up faster.
Please use the Options class to set preferences that you need and if you need to use an addon use the driver.install_addon("path/to/addon")
you can install selenium 4, which is in beta, via pip install selenium --pre
I noted in your code you were writing to user.js, which is a custom file for FireFox. Have you considered creating on these files manually outside of Gecko?
Also have you looked at mozprofile?
Thanks to source code provided in answer of Life is complex in link!. I have the chance to look through geckodriver source.
EXPLANATION
I believe that the reason you could not find out any rust_tmp in source because it is generated randomly by Profile::new() function.
When I look deeper in code structure, I saw that browser.rs is the place where the browser is actually loaded which is called through marionette.rs. If you noticing carefully, LocalBrowser::new method will be called whenever a new session is initialized and the profile will be loaded in that state also. Then by checking browser.rs file, there will be a block code line 60 - 70 used to actually generate profile for new session instance. Now, what need to do is modifying this path to load your custom profile.
SHORT ANSWER
Downloading zip file of geckodriver-0.30.0, extracting it by your prefer zip program :P
Looking on src/browser.rs of geckodriver source, in line 60 - 70, hoping you will see something like this:
let is_custom_profile = options.profile.is_some();
let mut profile = match options.profile {
Some(x) => x,
None => Profile::new()?,
};
Change it to your prefer folder ( hoping you know some rust code ), example:
/*
let mut profile = match options.profile {
Some(x) => x,
None => Profile::new()?,
};
*/
let path = std::path::Path::new("path-to-profile");
let mut profile = Profile::new_from_path(path)?;
Re-compile with prefer rust compiler, example:
Cargo build
NOTE
Hoping this info will help you someway. This is not comprehensive but hoping it is good enough hint for you like it is possible to write some extra code to load profile from env or pass from argument, it is possible but I'm not rust developer so too lazy for providing one in here.
The above solution is work fine for me and I could load and use directly my profile from that. Btw, I work on Archlinux with rust info: cargo 1.57.0.
TBH, this is first time I push comment on stackoverflow, so feel free to correct me if I'm wrong or produce unclear answer :P
Update
I worked in geckodriver 0.30.0 which will not be the same as geckodriver 0.29.1 mentioned by Life is complex. But the change between 2 versions just be split action, so the similar modify path in version 0.29.1 will be included in method MarionetteHandler::start_browser in file src/marionette.rs.
Since my starting point is Life is complex answer, please looking through it for more information.
I've come up with a solution that 1) works with Selenium 4.7.0--however, I don't see why it wouldn't work with 3.x as well, 2) allows the user to pass in an existing Firefox profile dynamically via an environment variable--and if this environment variable doesn't exist, simply acts "normally", and 3) if you do not want a temporary copy of the profile directory, simply do not pass the source profile directory to Selenium.
I downloaded Geckodriver 0.32.0 and made it so that you simply need to provide the Firefox profile directory via the environment variable FIREFOX_PROFILE_DIR. For example, in C#, before you create the FirefoxDriver, call:
Environment.SetEnvironmentVariable("FIREFOX_PROFILE_DIR", myProfileDir);
The change to Rust is in browser.rs, line 88, replacing:
let mut profile = match options.profile {
ProfileType::Named => None,
ProfileType::Path(x) => Some(x),
ProfileType::Temporary => Some(Profile::new(profile_root)?),
};
with:
let mut profile = if let Ok(profile_dir) = std::env::var("FIREFOX_PROFILE_DIR") {
Some(Profile::new_from_path(Path::new(&profile_dir))?)
} else {
match options.profile {
ProfileType::Named => None,
ProfileType::Path(x) => Some(x),
ProfileType::Temporary => Some(Profile::new(profile_root)?),
}
};
You may refer to my Git commit to see the diff against the original Geckodriver code.
The new driver by default creates a new profile if no options are set. To use a existing profile, one way to do this is to set the system property webdriver.firefox.profile before creating the firefox driver. A small code snippet that can create a firefox driver (given you have locations for geckodriver, and the firefox profile):
System.setProperty("webdriver.gecko.driver","path_to_gecko_driver");
System.setProperty("webdriver.firefox.profile", "path_to_firefox_profile");
WebDriver driver = new FirefoxDriver();
You could even set these system properties using the env. variables and skip defining them everywhere.
Another way to do this is to use the FirefoxOptions class which allows you to configure a lot of options. To start with, take a look at org.openqa.selenium.firefox.FirefoxDriver and org.openqa.selenium.firefox.FirefoxOptions. A small example:
FirefoxOptions options = new FirefoxOptions();
options.setProfile(new FirefoxProfile(new File("path_to_your_profile")));
WebDriver driver = new FirefoxDriver(options);
Hope this is helpful.
You can create firefox profile which will be clean and name it as SELENIUM
So When initializing the Webdriver get the profile which you have already created through the code, so that it wont create any new temp profiles all the time.
ProfilesIni allProfiles = new ProfilesIni();
FirefoxProfile desiredProfile = allProfiles.getProfile("SELENIUM");
WebDriver driver = new FirefoxDriver(desiredProfile);
That way, you assure that this profile will be used anytime you do the tests.
-Arjun
You can handle this by using --
FirefoxProfile profile = new FirefoxProfile(new File("D:\\Selenium Profile..."));
WebDriver driver = new FirefoxDriver(profile);
There is one more option but it inherits all the cookies, cache contents, etc. of the previous uses of the profile let’s see how it will be --
System.setProperty("webdriver.firefox.profile", "MySeleniumProfile");
WebDriver driver = new FirefoxDriver(...);
Hope this answers your question in short.
In my application I cannot set geckodriver executable location using System.setProperty and I cannot set it in the path.
Why? Because my app is multi-tenant... and each tenant has their own directory where Firefox and Geckodriver is copied and ran. This is due to bugs in the Firefox + Geckodriver, where infinite javascript loops and several other situations cause Firefox to hang until manual kill. Sometimes quit fails to kill things completely as well. So we need to supply a custom geckodriver location within the JVM per-tenant. Thus the problem.
So I am instead using:
driverService = new GeckoDriverService.Builder()
.usingDriverExecutable(new File(geckoDriverBinaryPath))
.build();
driverService.start();
RemoteWebDriver driver = new RemoteWebDriver(driverServiceUrl, capabilities);
But that is making me use a RemoteWebDriver when I am not remote.
Is there a better way to do this?
Rather than calling start() on the FirefoxDriverService object, why not simply use the FirefoxDriver constructor that takes the service?
driverService = new GeckoDriverService.Builder()
.usingDriverExecutable(new File(geckoDriverBinaryPath))
.build();
WebDriver driver = new FirefoxDriver(driverService);
As the questions stands it is still too broad. There are some unknowns: How are you running this? JUnit?, Maven?, Jenkins? And I am still not clear where this per-tenat geckoDriverBinaryPath comes from and how it is passed around.
What is wrong with just using:
System.setProperty("webdriver.gecko.driver", geckoDriverBinaryPath);
You could set an environment variable in your OS. Something like export geckoDriverBinary=/some/path and then in your code read it back using:
String geckoDriverBinaryPath = System.getenv("geckoDriverBinary");
System.setProperty("webdriver.gecko.driver", geckoDriverBinaryPath);
...
If you are running it from command line, either straight up or using Maven, you could pass the variable in like -DgeckoDriverBinaryPath=/some/path and then in your code read it back using:
String geckoDriverBinaryPath = System.getProperty("geckoDriverBinary");
System.setProperty("webdriver.gecko.driver", geckoDriverBinaryPath);
...
If the different tenants have the path fixed, you could write a utility function that would detect which tenant it is being run on, and set the property accordingly.
This answer is probably going to get closed as not-answer, but more of a discussion. :(
Opening a browser (google chrome) in a different language through selenium WebDriver works fine when running on PC, as described here. But when trying it on linux based systems, or mac-os, it simply doesn't work, and the browser opens on it's default language. I tried using different language code, such as "es_ES" or "es-ES" instead of "es", but nothing helped. Is it a different language code for linux, or is it a different way to manipulate the web driver and not use the "--lang" command?
Thanks.
I haven't try it but I think you can change the setting from chrome itself as :-
settings -> Lamguages -> Add languages.
Add your language there and give a try. remove other language if required.
For IE refer below link :-
http://www.reliply.org/info/internet/http-accept-lang.html
I have also found a code on same link you shared. Have you tried it?
DesiredCapabilities jsCapabilities = DesiredCapabilities.chrome();
ChromeOptions options = new ChromeOptions();
Map<String, Object> prefs = new HashMap<>();
prefs.put("intl.accept_languages", language);
options.setExperimentalOption("prefs", prefs);
jsCapabilities.setCapability(ChromeOptions.CAPABILITY, options);
Source :-
Set Chrome's language using Selenium ChromeDriver
Maybe you also need to set prefs > intl > accept_language: en-GB
"desiredCapabilities": {
"browserName": "chrome",
"chromeOptions": {
"args": ["--lang=en-GB"],
"prefs": {
"intl": {
"accept_languages": "en-GB"
}
}
}
}
As you can read at developer.chrome.com, there is a system-dependand way to set language for Chrome. On Linux, a environment variable is required.
I created a Bash script like this:
#!/bin/sh
LANGUAGE="en" "/home/plap/projects/pdf-exporter/chromedriver" $*
Then I use the path of the script in the place of the path of the real chromedriver executable.
Moreover, since I need to swich language programmatically, I made code to save scripts like that programmatically at each new language required. Indeed the code calls:
System.setProperty("webdriver.chrome.driver", executableFile.getAbsolutePath());
in a synchronized block together with new ChromeDriver(options)... Yes, it's horrible!
Consider a simple DefaultSelenium object
DefaultSelenium sel = new
DefaultSelenium("http://localhost:8080/myapp",4444,"*iexplore","/myAppLevel1");
Now my server is set with the option of -forcedBrowserMode "*firefox" in the command line when I start it up. However, I have 2 different batch files to start the Server, one forced in firefox, one forced to IE. FYI, the -forcedBrowserMode overrides the settings within of the instantiated java object.
The problem is from java, I can't seem to find a way to determine which browser my DefaultSelenium object is running on... I was thinking something like:
sel.getBrowserName();
But nothing like that exists. Are there any other creative ways of doing this?
I need to know because with a GWT web application, to click on a button you need to do it differently based on the browser. As well, you may wonder why I even use the -forcedBrowserMode, because then I can use custom setup firefox/ie installs to test on.
Thanks in advance for help!
I think that you can get the browser by executing some JavaScript, for example verify navigator.userAgent or any browser specific object, for example document.defaultView will be null in IE and not null in FF, something like this:
DefaultSelenium sel = ...
String res = sel.getEval("document.defaultView ? false : true");
boolean isIE = "true".equals(res);
What is the experience of working with OpenOffice in server mode? I know OpenOffice is not multithreaded and now I need to use its services in our server.
What can I do to overcome this problem?
I'm using Java.
With the current version of JODConverter (3.0-SNAPSHOT), it's quite easy to handle multiple threads of OOo in headless-mode, as the library now supports starting up several instances and keeping them in a pool, by just providing several port numbers or named pipes when constructing a OfficeManager instance:
final OfficeManager om = new DefaultOfficeManagerConfiguration()
.setOfficeHome("/usr/lib/openoffice")
.setPortNumbers(8100, 8101, 8102, 8103)
.buildOfficeManager();
om.start();
You can then us the library e.g. for converting documents without having to deal with the pool of OOo instances in the background:
OfficeDocumentConverter converter = new OfficeDocumentConverter(om);
converter.convert(new File("src/test/resources/test.odt"), new File("target/test.pdf"));
Yes, I am using OpenOffice as a document conversion server.
Unfortunately, the solution to your problem is to spawn a pool of OpenOffice processes.
The commons-pool branch of JODConverter (before it moved to code.google.com) implemented this out-of-the-box for you.
Thanks Bastian. I found another way, based on Bastian's answer. Opening several ports it provides access to create multithreads. But without many ports(enought several) we can improve performence by increase task queue timeout here is a documentation. And one thing again, we decided not to start and stop officeManager on each convertion process.At the end, I solved this task by this approach:
public class JODConverter {
private static volatile OfficeManager officeManager;
private static volatile OfficeDocumentConverter converter;
public static void startOfficeManager(){
try {
officeManager = new DefaultOfficeManagerConfiguration()
.setOfficeHome(new File('libre office home path'))
.setPortNumbers(8100, 8101, 8102, 8103, 8104 )
.setTaskExecutionTimeout(600000L) // for big files
.setTaskQueueTimeout(200000L) // wait if all port were busy
.buildOfficeManager();
officeManager.start();
// 2) Create JODConverter converter
converter = new OfficeDocumentConverter(officeManager);
} catch (Throwable e){
e.printStackTrace();
}
}
public static void convertPDF(File inputFile, File outputFile) throws Throwable {
converter.convert(inputFile, outputFile);
}
public static void stopOfficeManager(){
officeManager.stop();
}
}
I call JODConverter's convertPDF when convertion is need. It will be stopped only when application was down.
OpenOffice can be used in headless mode, but it has not been built to handle a lot of requests in a stressfull production environment.
Using OpenOffice in headless mode has several issues:
The process might die/become unavailable.
There are several memory leaks issues.
Opening several OpenOffice "workers" does not scale as expected, and needs some tweaking to really have different open proccesses (having several OpenOffice copies, several services, running under different users.)
As suggested, jodconverter can be used to access the OpenOffice process.
http://code.google.com/p/jodconverter/wiki/GettingStarted
you can try this:
http://www.jopendocument.org/
its an opensource java based library that allows you to work with open office documents without open office, thus removing the need for the OOserver.
Vlad is correct about having to run multiple instances of OpenOffice on different ports.
I'd just like to add that OpenOffice doesn't seem to be stable. We run 10 instances of it in a production environment and set the code up to re-try with another instance if the first attempt fails. This way when one of the OpenOffice servers crashes (or doesn't crash but doesn't respond either) production is not affected. Since it's a pain to keep restarting the servers on a daily basis, we're slowly converting all our documents to JasperReports (see iReport for details). I'm not sure how you're using the OpenOffice server; we use it for mail merging (filling out forms for customers). If you need to convert things to PDF, I'd recommend iText.