我在所有主流浏览器中测试了以下代码，包括老式的explorer(通过兼容模式)，它在任何地方都能正常工作:

$filename = $_GET['file']; //this string from $_GET is already decoded
if (strstr($_SERVER['HTTP_USER_AGENT'],"MSIE"))
  $filename = rawurlencode($filename);
header('Content-Disposition: attachment; filename="'.$filename.'"');

我最终在“download.php”脚本中编写了以下代码(基于这篇博文和这些测试用例)。

$il1_filename = utf8_decode($filename);
$to_underscore = "\"\\#*;:|<>/?";
$safe_filename = strtr($il1_filename, $to_underscore, str_repeat("_", strlen($to_underscore)));

header("Content-Disposition: attachment; filename=\"$safe_filename\""
.( $safe_filename === $filename ? "" : "; filename*=UTF-8''".rawurlencode($filename) ));

只要只使用iso-latin1和“safe”字符，就使用标准的filename="…";如果不是，它会添加文件名*=UTF-8 " url编码的方式。根据这个具体的测试用例，它应该从MSIE9起，并在最近的FF, Chrome, Safari;在较低的MSIE版本中，它应该提供包含ISO8859-1版本的文件名，在非此编码的字符上使用下划线。

最后注意:最大值。在apache上，每个报头字段的大小为8190字节。UTF-8每个字符最多可以有四个字节;在rawurlencode之后，每个字符是x3 = 12字节。非常低效，但理论上仍然可以在文件名中有超过600个“smiles”%F0%9F%98%81。

将文件名放在双引号中。帮我解决了问题。是这样的:

Content-Disposition: attachment; filename="My Report.doc"

http://kb.mozillazine.org/Filenames_with_spaces_are_truncated_upon_download

我测试了多种选择。浏览器不支持这些规格，并且表现不同，我相信双引号是最好的选择。

库类Unicode中的方法mimeHeaderEncode($string)可以完成这项工作。

$file_name= Unicode::mimeHeaderEncode($file_name);

drupal/php中的例子:

https://github.com/drupal/core-utility/blob/8.8.x/Unicode.php

/**
   * Encodes MIME/HTTP headers that contain incorrectly encoded characters.
   *
   * For example, Unicode::mimeHeaderEncode('tést.txt') returns
   * "=?UTF-8?B?dMOpc3QudHh0?=".
   *
   * See http://www.rfc-editor.org/rfc/rfc2047.txt for more information.
   *
   * Notes:
   * - Only encode strings that contain non-ASCII characters.
   * - We progressively cut-off a chunk with self::truncateBytes(). This ensures
   *   each chunk starts and ends on a character boundary.
   * - Using \n as the chunk separator may cause problems on some systems and
   *   may have to be changed to \r\n or \r.
   *
   * @param string $string
   *   The header to encode.
   * @param bool $shorten
   *   If TRUE, only return the first chunk of a multi-chunk encoded string.
   *
   * @return string
   *   The mime-encoded header.
   */
  public static function mimeHeaderEncode($string, $shorten = FALSE) {
    if (preg_match('/[^\x20-\x7E]/', $string)) {
      // floor((75 - strlen("=?UTF-8?B??=")) * 0.75);
      $chunk_size = 47;
      $len = strlen($string);
      $output = '';
      while ($len > 0) {
        $chunk = static::truncateBytes($string, $chunk_size);
        $output .= ' =?UTF-8?B?' . base64_encode($chunk) . "?=\n";
        if ($shorten) {
          break;
        }
        $c = strlen($chunk);
        $string = substr($string, $c);
        $len -= $c;
      }
      return trim($output);
    }
    return $string;
  }

我使用以下代码片段进行编码(假设fileName包含文件的文件名和扩展名，即:test.txt):

PHP:

if ( strpos ( $_SERVER [ 'HTTP_USER_AGENT' ], "MSIE" ) > 0 )
{
     header ( 'Content-Disposition: attachment; filename="' . rawurlencode ( $fileName ) . '"' );
}
else
{
     header( 'Content-Disposition: attachment; filename*=UTF-8\'\'' . rawurlencode ( $fileName ) );
}

Java:

fileName = request.getHeader ( "user-agent" ).contains ( "MSIE" ) ? URLEncoder.encode ( fileName, "utf-8") : MimeUtility.encodeWord ( fileName );
response.setHeader ( "Content-disposition", "attachment; filename=\"" + fileName + "\"");

以下文件链接自Jim在回答中提到的RFC草案，进一步解决了这个问题，在这里值得直接注意:

HTTP内容处理头和rfc2231 /2047编码的测试用例