The Windows localized installers are not honoring the CHARSET/FONTSIZE/FONTNAME directives from install.ini. This means that in many cases improper fonts are being used to display the installer dialogs. The directives can be found here: http://lxr.mozilla.org/aviarybranch/source/toolkit/locales/en-US/installer/windows/install.it#2 However, they don't affect anything useful. Instead, the fonts are taken from the setup resource DLL http://lxr.mozilla.org/aviarybranch/source/toolkit/mozapps/installer/windows/wizard/setuprsc/setuprsc.rc#60 The Japanese team has been hacking around this problem by using a resource editor to change the font name prior to release. But this obviously sucks for automated releases. I *think* that a simple solution involves sending a WM_SETFONT message when we initialize each dialog, but I'm not sure exactly where this call should be made; I feel like I'm wandering through a spaghetti of win32 API, so if anyone else understands this code better than I, please step forward. An alternative hack is to rebuild setuprsrc.dll for each locale, and use preprocessor defines to hardcode the font settings into the resource DLL. It might be easier, but it kinda sucks. As a sidenote that might also need fixing, the title font is hardcoded to Trebuchet MS Bold at http://lxr.mozilla.org/aviarybranch/source/toolkit/mozapps/installer/windows/wizard/setup/dialogs.c#226 This should be configurable as well.
Maybe "MS Shell Dlg" is what you want in the resource file, see http://msdn.microsoft.com/library/default.asp?url=/library/en-us/intl/nls_2np0.asp
Ben Goodger says he can't reproduce this bug. What does an instance of this bug look like? Could someone who's seeing it attach a screenshot?
Created attachment 164460 [details] Screenshot of ja-JP installer without fix on WinXP. This is the screenshot taken with l10n nightly (as is) on Windows XP. "Developer Tools" and "Quality Feedback Agent" labels of checkbox have wrong code. # of cource they are translated in japanese.
Created attachment 164462 [details] Screenshot of ja-JP installer with fix on WinXP. Screenshot taken with rc1 (fixed by hand using reshacker) on Windows XP. Whole text are good including the checkbox labels.
Created attachment 164463 [details] Screenshot of ja-JP installer without fix on Win98. This is the screenshot taken with l10n nightly (as is) on Windows 98. Whole japanese text will be a black bar...
In the installer, each dialog body use the setting written in setuprsrc.dll. Installer in ja-JP shows funny text because japanese text are shown with non-japanese font. If we change the value of setuprsrc.dll, text will be shown well # [FONT 8, "MS Sans Serif"] -> [FONT 10, "MS UI Gothic"] # "MS UI Gothic" is windows default japanese font for UI. When the specified font don't have the shape data for the text... [WinNT/2K/XP] just ignore the specified font and use the correct one. [Win95/98/ME] use the specified font and text will be funny. So, whole of installer dialog have funny text in win9x. It seems that WinXP sometimes cannot select correct font and make funny text partially like the first screenshot. # Installer have funny text only in the component selecting dialog on WinXP. The setting written in install.ini is used only in the preparing messege dialog which is whown before "Welcome to Mozilla Firefox" dialog. c.f. Only the body (main part) of dialog use the setting written in setuprsrc.dll. Frame part of dialog don't use the setting but use the OS setting UI font. # in the thrid screenshot, "Welcome to Mozilla Firefox" is whown with japanese # font on win98. "MS Shell Dlg" works well in win2K, XP (not tested with win9x) but our peer said that it seems to have some problem. # I'm asking about the detail now...
Created attachment 164479 [details] Screenshot of the dialog before "Welcome to MozFF" on WinXP This is the dialog shown at the start of the installer. This is taken with the ja-JP installer without fix on WinXP. install.ini setting was: FONTNAME=MS UI Gothic FONTSIZE=14 # to check where the value of FONTSIZE is used, I setted large font...
This is the workaround we have been used for ja-JP installer: 1. with 7-zip utility, extract installer of nightly build. 2. with reshacker, open the file 'setuprsc.dll'. resource hacker we use: http://www.users.on.net/johnson/resourcehacker/ 3. edit only hard-corded font setting. editting screenshot: http://skillup.jp/firefox/screenshot/reshacker01.png 4. save (re-compile) the file. 5. repeat step 2~4 for UninstallFirefox.exe too (in UninstallFirefox.zip). 6. re-package installer as written in the mozilla.org document. http://www.mozilla.org/projects/firefox/l10n/localize-release.html Only this is the change of our re-packaged installer. ant build script for re-packaging (I use this on WinXP): http://skillup.jp/firefox/makeinstaller.zip # this script will do last step only (step 6 above). sample of the changed file http://skillup.jp/firefox/installer/MS_UI_Gothic_9/setuprsc.dll http://skillup.jp/firefox/installer/MS_UI_Gothic_9/UninstallFirefox.exe # based on the 2nd round l10n build for 1.0 final. # - FONT 8, "MS Sans Serif" # + FONT 9, "MS UI Gothic"
This creates logistical nightmares when doing releases since we have to deal with sending modified installers around, waiting for people to make appropriate edits, etc., so marking blocking-aviary1.1+.
> "MS Shell Dlg" works well in win2K, XP (not tested with win9x) but our peer said > that it seems to have some problem. > # I'm asking about the detail now... Can you elaborate on the problem caused by "MS Shell Dlg" ?
Assignee: benjamin → masayuki
Created attachment 177961 [details] [diff] [review] Patch rv1.0
Attachment #177961 - Flags: review?(benjamin)
Priority: -- → P1
Version: 1.0 Branch → unspecified
The same problem of seamonky is bug 286670.
Attachment #177961 - Flags: review?(benjamin) → review+
Attachment #177961 - Flags: superreview?(dveditz)
Target Milestone: --- → Firefox1.1
Comment on attachment 177961 [details] [diff] [review] Patch rv1.0 didn't I review this already? sr=dveditz
Attachment #177961 - Flags: superreview?(dveditz) → superreview+
Comment on attachment 177961 [details] [diff] [review] Patch rv1.0 Japanese l10n team hopes that this patch is checked-in to 1.0.1 aviary branch. This patch's risk is low. But this problem is important for localization team.
Attachment #177961 - Flags: approval-aviary1.0.4?
(In reply to comment #14) > (From update of attachment 177961 [details] [diff] [review] ) > didn't I review this already? > > sr=dveditz > No. Maybe, it is bug 288670 that is seamonkey installer.
Comment on attachment 177961 [details] [diff] [review] Patch rv1.0 The risk is low. See comment 15.
Attachment #177961 - Flags: approval-aviary1.1a?
Comment on attachment 177961 [details] [diff] [review] Patch rv1.0 a=asa for checkin to the trunk for 1.1a
Attachment #177961 - Flags: approval-aviary1.1a? → approval-aviary1.1a+
checked-in to trunk.
We took the seamonkey equivalent for 1.7.7, nominating
(In reply to comment #20) > We took the seamonkey equivalent for 1.7.7, nominating This bug is seeking approval for a much larger patch which adds (or readds?) general support for other fonts in the installer. The spot fix we took on the 1.7 branch was a 2-line fix: http://bonsai.mozilla.org/cvsquery.cgi?treeid=default&module=MozillaTinderboxAll&branch=MOZILLA_1_7_BRANCH&branchtype=match&dir=&file=&filetype=match&who=&whotype=match&sortby=Date&hours=2&date=explicit&mindate=2005-03-23&maxdate=2005-03-25&cvsroot=%2Fcvsroot. Since the patch is large, how safe is it? Is it just readding code that was at one time on the 1.7 or aviary1.0 branches?
(In reply to comment #21) > Since the patch is large, how safe is it? Is it just readding code that was at > one time on the 1.7 or aviary1.0 branches? The risk is same as seamonkey's patch. It is low. Firefox and thunderbird installer ignore font settings on every text label. Therefore this patch has many lines. In case of seamonkey, the problem is two labels only. Therefore seamonkey's patch is two lines only. In ohter words, number of lines in patch is same as number of having problem labels.
The only benefit of this patch is that the Japanese installer bits will not need manual resouce-hack munging. It's only vaguely risky, so it should be up to Chase IMO, balancing the risk against the pains of manual repackaging.
Comment on attachment 177961 [details] [diff] [review] Patch rv1.0 Now that Firefox 1.0.3 has been released and the pressure is off to constantly have the branch in a can-ship state, we have a great opportunity to try this patch out there. You have approval to land this patch there. If it destabilizes the branch with no quick fix for that available, you'll have to back the patch out. Please help us mobilize build testing of various locales after the patch lands to ensure that no regressions exist as a result.
Attachment #177961 - Flags: approval-aviary1.0.4? → approval-aviary1.0.4+
Checking in toolkit/mozapps/installer/windows/wizard/setup/dialogs.c; /cvsroot/mozilla/toolkit/mozapps/installer/windows/wizard/setup/dialogs.c,v <-- dialogs.c new revision: 22.214.171.124.126.96.36.199; previous revision: 188.8.131.52.184.108.40.206 done connect: Connection refused at /cvsroot/CVSROOT/dolog.pl line 244.
Status: ASSIGNED → RESOLVED
Last Resolved: 14 years ago
Resolution: --- → FIXED
I found a regression that is bug 291143. I backed out the patch from 1.0.1 aviary branch.
Version: unspecified → Trunk
Existing patch backed out, not blocking a security release
Flags: blocking-aviary1.0.5? → blocking-aviary1.0.5-
This patch caused a regression when it landed on the aviary1.0.1 branch. Has a fix for that regression been developed? If so, I'd like for us to consider this patch again for a 1.0.x release if it means we can land a modified version of this patch that fixes the regression it caused, as well. Also, this patch landed on the trunk before the mozilla1.8 branch was created so it should be on that branch, as well. Masayuki, can you take a look at the ja-JP win32 installer that is being built by the l10n Windows build system and ensure that this is fixed in the post-1.5b1 packages?
(In reply to comment #30) > See bug 291143. We need to change intl resouce. > And we need bug 291909 too. Do you have cycles to create a new patch that contains the original patch for aviary1.0.1 along with the fixes it needs to land (addressing the regressions mentioned in bug 291143 and bug 291909)? I cannot guarantee I can get you an audience in a 1.0.x release post-1.0.7 but it would help my case if we had a patch in-hand, verified in advance, and ready to land.
> Masayuki, can you take a look at the > ja-JP win32 installer that is being built by the l10n Windows build system and > ensure that this is fixed in the post-1.5b1 packages? I don't know all problems are gone. The all text length is different by language. I don't know current static text objects have enough size for all language.
(In reply to comment #32) > I don't know all problems are gone. The all text length is different by language. > I don't know current static text objects have enough size for all language. Since I am recommending to the org we only redistribute ja-JP win32 installers that are built by our build systems locally, I suggest any problems with the ja-JP win32 installers be addressed ASAP. What we have in the mozilla1.8 nightlies is what we're going to ship in the release.
O.K. I will create the patch for 1.0.x. But I'm already working some regressions on 1.8 branch. I need have a time. Please wait few days.
Oops... I find a bug. -> REOPEN.
Status: RESOLVED → REOPENED
Resolution: FIXED → ---
Status: REOPENED → ASSIGNED
The new patch and for 1.0.x patch will come tomorrow.
Created attachment 197267 [details] [diff] [review] Patch for Trunk/1.8branch The 3 windows are not modified font settings. 1. The license field. 2. The install path of selecting path dialog. 3. The install path of installing options summary dialog.
Attachment #164460 - Attachment is obsolete: true
Attachment #164462 - Attachment is obsolete: true
Attachment #164463 - Attachment is obsolete: true
Attachment #164479 - Attachment is obsolete: true
Attachment #177961 - Attachment is obsolete: true
Attachment #181055 - Attachment is obsolete: true
Attachment #197267 - Flags: review?(benjamin)
Created attachment 197268 [details] [diff] [review] Patch for 1.0.x branch This patch marged attachment 177961 [details] [diff] [review](this bug), attachment 181425 [details] [diff] [review](bug 291143), attachment 181851 [details] [diff] [review](bug 291909) and previous patch(attachment 197267 [details] [diff] [review], this bug).
Attachment #197268 - Flags: review?(benjamin)
Whiteboard: [needs review benjamin]
Comment on attachment 197268 [details] [diff] [review] Patch for 1.0.x branch Chase, this is a pretty large change for the 1.0.x branch, and it will require changes to all the locales' install.it; I'm not sure it's worth it.
Attachment #197268 - Flags: review?(benjamin) → review+
Comment on attachment 197267 [details] [diff] [review] Patch for Trunk/1.8branch oops, I did not mean to mark approval+
Attachment #197267 - Flags: approval1.8b5+ → approval1.8b5?
checked-in to trunk.
Status: ASSIGNED → RESOLVED
Last Resolved: 14 years ago → 13 years ago
Resolution: --- → FIXED
Whiteboard: [needs review benjamin] → [needs approval]
Comment on attachment 197267 [details] [diff] [review] Patch for Trunk/1.8branch Approved per 9/26 bug triage meeting.
Attachment #197267 - Flags: approval1.8b5? → approval1.8b5+
Please get the 1.8branch version of the patch checked in on the branch ASAP.
Oh, sorry. I forgot this check-in. But I will sleep soon. I will can check-in after 14 hours. If anyone can check-in the patch, please check-in to 1.8branch.
Whiteboard: [needs approval] → Should we land on 1.0.x?
checked-in to 1.8 branch.
out of scope for the 1.0 branch, approval denied per drivers. New customers will get 1.5, old customers have already seen this brokenness in earlier 1.0 versions.
Flags: blocking-aviary1.0.8? → blocking-aviary1.0.8-
Status: RESOLVED → VERIFIED
QA Contact: bugzilla → installer
You need to log in before you can comment on or make changes to this bug.